The History of the Critical Path Method

By Victor Hoang · Updated Oct 23, 2026

Picture DuPont's engineering department in late 1956. Design, procurement and construction each did their own planning and scheduling. They worked largely independent of one another. The authors of the 1959 paper say that led to a lack of coordination.

Morgan R. Walker of the DuPont engineering department and James E. Kelley Jr., a mathematician at Remington Rand UNIVAC, built the critical path method (CPM) between December 1956 and February 1959. Its first computer runs came in 1957 on a UNIVAC I. It reached print in the 1959 Eastern Joint Computer Conference proceedings. The through-line is an import-and-transform story. An engineering company's scheduling problem became a linear program, the linear program became a picture, and the picture outlived the math that produced it.

In their 1959 paper, Kelley and Walker say the company's Integrated Engineering Control Group started a survey in late 1956 to find out whether a computer could prepare a master schedule and revise it as conditions changed. This article follows what came out of that survey, who else was working on the same problem, and how the method ended up inside software and courtrooms. For how the method works today, see the critical path method. For the wider field, see the history of project management.

Timeline at a Glance

Dates separate the working method from the first publication.

The Bar Chart Problem (1896-1956)

Charts show when work happens, not what it waits for

Start with paper strips. In 1896 the Polish engineer Karol Adamiecki built a scheduling chart from scaled paper strips, one per activity, with each activity's predecessors listed above it, as Patrick Weaver describes it. Our piece on Adamiecki and the harmonogram tells why almost nobody outside Poland saw it. Henry Gantt's charts, which came to dominate the English-speaking world, did a different job. Weaver argues they were retrospective production records, and Gantt built them for repetitive factory work. See the history of the Gantt chart for that split. Neither chart was designed to answer the question a project manager asks first: if this job slips, what moves?

Sequencing itself was far older than either chart. Builders ran ancient megaprojects with logbooks and work gangs, but a logbook records what happened. It can't calculate what a delay will do next. What was missing was a way to write the order of work down so that a machine could check it. Jerome Wiest, writing in 1977, says the network methods had their roots in flow diagrams, Gantt charts, the line-of-balance technique and the milestone method. An American Institute of Constructors review adds that Melvin Salveson described a "precedence graph" for balancing assembly lines in 1955, and also used the term "slack time". Kelley himself later called the network idea obvious to any mathematician of the 1950s, since project sequences resemble the structures studied in graph theory and set theory.

So the ingredients were on the table. What nobody had done was combine them with a computer and a real engineering budget.

What this means for teams today: a schedule that lists only start and finish dates hides the real question, which is order. If your plan can't answer "what waits on this task?", it is still a bar chart, whatever the software calls it.

The DuPont Experiment (1956-1959)

A chemical company asks whether a computer can schedule

In 1957 DuPont had a UNIVAC I in Newark, Delaware, and Remington Rand would supply the programs that ran on it. The man given the scheduling problem was Morgan Walker, who joined the Integrated Engineering Control Group around Christmas 1956. Walker's own account says the group had been set up that autumn by Chief Engineer Granville Read, whose engineering department employed over 3,000 permanent people, to look for better ways of doing engineering work. He remembers being told only to see what could be done about scheduling, and being warned by friends that taking the job was political suicide.

The first move was already on paper. On December 20, 1956, two DuPont mathematicians, Philip Hayward and John Robinson, proposed in a memo that the UNIVAC I be tied into construction scheduling. DuPont did not build on it, because their group was self-supporting and would have needed funding from outside the department to continue. Walker turned to John Mauchly's applications group at Remington Rand UNIVAC, and in early January 1957 Mauchly introduced him to Kelley. Kelley had a road-building paper due at a Case Institute conference at the end of the month, and the first Interstate highway bill had been signed into law the previous June, so Kelley saw scheduling as a way to widen the market for computers.

Kelley's contribution was a model. He treated each activity as having a normal duration and cost and a faster "crash" duration at higher cost, then asked how to shorten the whole project at the least added cost. The answer was a parametric linear program, and his 1961 paper in Operations Research later set out the mathematics. A write-up of the method existed by April 25, 1957, and on May 7 a meeting in Newark turned it into a joint project. Walker and Kelley reconstructed the cost afterward at roughly 1,442 DuPont man-days and 400 Remington Rand man-days, about $226,000 in 1956 dollars, and they flag it as an estimate.

The first test network is the best story. Walker and James Sage mapped a small mixing and packaging plant that the engineering group had developed the year before, which they named the Fisher Works after Walker's boss, George Fisher. The result was 61 jobs, 8 timing restraints and 16 dummies. The dummies came from a blackboard session on July 3, 1957, when Walker brought the logic problems he was having and he and Kelley solved them by inventing the dummy arrow. Walker says the exercise also made planners decide a strategy for the whole job first, for example whether to erect the tank farm with the building or after it. In his words, CPM forced the issue.

The arrow diagram itself came late, and for a practical reason. Weaver quotes a 1983 interview in which Kelley says the mathematics came first and he drew the arrows afterward because DuPont's managers could not follow the algorithm. In the joint memoir, Kelley describes the two ways to draw a network. Cards for jobs with arrows between them give the node form. Arrows for jobs, with dummies where the logic needs them, give the CPM form. He calls the two mathematically equivalent.

The live tests

The 1959 paper records what happened next. The September 1957 demonstration showed promise, and by late December a team of six engineers had been formed and given a 40-hour workshop on the method. Their test, "Project A," was a new chemical plant facility capitalized at $10,000,000. The team worked separately from DuPont's normal schedulers, and their schedules were not used to run the project. They analyzed only the construction steps, from the point where design was about 30 percent complete, and the network grew to 393 jobs, 156 design and delivery restraints and 297 dummies.

The findings were what sold it. Of 156 deliveries, only seven turned out to be critical, and three of those were missing from the list the normal scheduling group had prepared. The computer showed two months could be gained at no added cost, and two more for about a 1 percent rise in variable direct cost, because only about 10 percent of the jobs were critical. After about 40 percent of the plan changed in 1958, updating took about a tenth of the effort the original had. The team also went into far more detail than anyone expected, which made the UNIVAC I programs impractical, so Remington Rand reprogrammed the system for the larger 1103A and later prepared programs for the 1105.

The cleanest result came from a maintenance shutdown, not from construction. DuPont's Louisville Works makes an intermediate in the neoprene process, a self-detonating material, so all maintenance had to happen while a unit was down. By March 1959 the test was complete. The paper reports that the average shutdown fell from 125 hours to 93, with 78 hours projected, and that this one application paid back the development cost five times over. The authors also found that only about 10 percent of shutdown jobs were critical, so surprise work could be absorbed in float.

The parallel strand: PERT

At the same moment, a second network method was taking shape inside the Navy. The Special Projects Office, which ran the Polaris missile program through about 3,000 contractors and agencies, set up a team of operations researchers from Booz, Allen & Hamilton, Lockheed Missile Systems Division and the Special Projects Office in early 1958. They asked experts for three durations per activity (optimistic, most likely, pessimistic), fitted a probability distribution, and ran the sums on the NORC computer at Dahlgren, Virginia. The authors, Malcolm, Roseboom, Clark and Fazar, described the result in Operations Research in 1959. For the mechanics, see what a PERT chart is.

The two teams say they never met. Kelley recalls that he and Walker had not heard of PERT until early 1959, and the earliest meeting they can document is April 2, 1959. An AIC Journal review of the record notes that Harvey Sapolsky's 1972 study of Polaris found an acrimonious dispute over parentage once PERT became famous. The vocabulary crossed anyway. Weaver reports that the PERT team coined the term "critical path," and that Kelley and Walker had called the same chain the main chain. Their own memoir says they adopted the term for marketing around May 1959 and named their product the Critical-Path Method. This strand overlaps the DuPont work, and PERT's story runs on through the 1960s in the next sections. For the wider question of credit, see who invented project management.

The paper makes two points that shaped everything after it. Kelley and Walker say their method rests on technology and sequence alone, so considerations of manpower and equipment are "conspicuous by their absence". And they predict uses far beyond chemical plants, from highways and dams to retooling programs.

What this means for teams today: Kelley later argued that drawing the network and calculating early and late times and float gives about 90 percent of the value of network methods. Do that first, and treat everything beyond it as optional.

The Mauchly Years (1959-1963)

A computer maker's side project becomes a consulting boom

Consider the first public workshop, held at the Franklin Institute in Philadelphia on November 16-20, 1959. Walker recalls that only five people had signed up a week before, with 20 hoped for, so he spent the expected profit on long-distance calls to Fortune 500 engineering departments, and 15 had signed up by Friday afternoon. That is how CPM left DuPont. Mauchly's group at Remington Rand was disbanded in early 1959, and Mauchly started Mauchly Associates, moving into offices in Ambler, Pennsylvania, on April 9. Walker joined that July, after Kelley. Walker says DuPont's own engineering department lost interest once Read retired, and did not adopt CPM as standard practice until 1968.

Mauchly Associates kept the method alive, and the era runs alongside the PERT strand described above. Kelley and Walker presented their paper at the Eastern Joint Computer Conference in Boston on December 1-3, 1959, and were classed as professionals and disqualified from the best-paper award. Weaver says the firm simplified the process to focus on schedule rather than cost. Their consulting work also added a missing piece: Walker, with Charles Bachman and Jack Westley of Dow Chemical, devised a resource-leveling method in early 1960, because a Dow plan called for about 150 boilermakers when only about 50 were in the area. By the end of 1962 the firm had trained close to 1,000 people. Weaver reports that Catalytic Construction of Philadelphia was the first large construction company to adopt CPM, in 1961.

Money and mandates did the rest. Wiest writes that in the early 1960s the U.S. government began requiring network reporting on certain contracts, mostly in defense research and development, so the question of which method came first became moot. PERT/COST was added under Defense Department and NASA sponsorship in 1962, the step that opens the history of earned value management. Training courses, books and "instant experts" followed. Weaver notes that CPM was popular but expensive, and that solving some scheduling problems could cost about the price of a small car.

The new problem was hype. Wiest's account describes managers forced to adopt PERT, "wall-to-wall PERT" charts, mountains of computer printout, and companies that kept a second system to run the work while the network fed the government. Walker says interest peaked early and fell to a low ebb by the late 1960s, partly because of the downturn in aerospace, before recovering in the early 1970s, especially in Europe.

What this means for teams today: when a customer or a regulator requires a method, the paperwork spreads faster than the understanding. Ask who reads the network before you fund anyone to maintain it.

The Precedence Turn (1958-1970s)

A manual method from Stanford outlasts the arrow

This strand starts in the same year as PERT and runs alongside the Mauchly years. At Stanford in 1958, John Fondahl was working with circles and connecting lines. Wiest says Fondahl developed a "circle-and-connecting-line" diagram in 1958, and that at the same time Bernard Roy in France began work on a similar activity-on-node method of potentials. Weaver adds that Fondahl's Stanford team was commissioned by the Navy's Bureau of Yards and Docks in 1958, and that his report on a non-computer approach to CPM for construction was published in 1961. Walker dates it to November 1961. Weaver says it sold over 20,000 copies.

The idea moved from paper to mainframe through a contractor. The H.B. Zachry Company of San Antonio had been experimenting with CPM since 1959, and Weaver reports it built a precedence version in 1962. In March 1963 Zachry and IBM entered a joint venture. In early 1964 IBM announced its Project Control System for the 1440 computer, which used the name "precedence diagramming," and the Stanford team adopted the same name. It was also the first computerized system to include lags for overlapping activities. Meanwhile Moder and Phillips had proposed the term "activity-on-node" in a 1964 book.

Walker was not a fan. He found the clutter of connector arrows confusing on large networks, and he notes that some commercial packages convert node networks to the arrow form before calculating. The market decided otherwise. Weaver writes that IBM's dominance made precedence diagramming the dominant mainframe method, and that by 2004 it was virtually the only system in general use, with PERT gone and the arrow form rarely seen outside academic papers.

CPM itself did not fade the way the hype did. Wiest cites a 1974 survey of the 400 largest U.S. construction firms in which over 80 percent of 235 respondents used critical path methods, up from 47 percent before 1965. He also cites Norman Gaither's survey of manufacturing firms in seven south-central states, in which PERT and CPM ranked first among operations research techniques, used by 33.5 and 32.4 percent of firms. Not every verdict was kind. David Murphy's NASA study, quoted by Wiest, found the techniques over-used and over-detailed in some cases. Wiest also notes that PERT was not part of the vocabulary of the Apollo managers.

What this means for teams today: the picture that wins is the one the cheapest tool can draw. The Fondahl form survived because it could be sketched by hand and then coded, not because it beat the arrow on elegance.

The Desktop Years (1980-1990s)

Scheduling leaves the specialists and lands on every PC

Before the early 1980s, Weaver says, you built a schedule on a mini or mainframe, drew and calculated it by hand, or did both, checking the manual results first because computer time cost too much to waste on errors. Schedulers learned their trade through apprenticeship. Then the machines shrank. Micro Planner v1.0 ran on the Apple II in 1980, "The Planner" shipped for the IBM PC in 1982, and Primavera was founded in May 1983. The Microsoft Project Users Group dates Microsoft Project version 1 to 1984, for DOS, with Microsoft buying all rights in 1985.

The 1990s followed the same path with cheaper tools. Weaver says the arrival of inexpensive PC schedulers meant that by the early 1990s almost nobody scheduled manually, and the number of people creating schedules part-time and untrained exploded. He calls the result a myth that anyone can schedule, and he says many schedules are bar charts masquerading as CPM schedules. A 2012 review of the Gantt chart's history finds that the network methods did not displace the bar chart. Instead, the two combined, with the network doing the calculating and the bar chart showing the answer. For that thread, see the history of the Gantt chart.

What this means for teams today: a tool that calculates a critical path for you does not check your logic. Missing links and padded durations still produce a confident wrong answer.

The Evidence Years (1990s-present)

The schedule becomes a record, and a target of reform

This era starts as the desktop boom peaks. Weaver observes that once schedulers had been replaced by desktop users, the professional schedulers who remained were mostly consultants, particularly claims consultants. That tells you where the method's value had moved. A CPM schedule is a dated statement of intent, and when a project runs late the logic and updates become evidence of who caused the delay. The Society of Construction Law's Delay and Disruption Protocol, second edition February 2017, says that in most cases the contractor's programme should be based on the critical path method, and that an extension of time is generally due where an employer risk event impacts the critical path. Some of its delay analyses recalculate a logic-linked baseline using CPM software.

Critics came from inside the method's own logic. In 1997 Eliyahu Goldratt's business novel Critical Chain argued that safety hidden in each task estimate gets wasted, and proposed pooling it into buffers. A PMI literature review notes that Goldratt defines the critical chain as the longest chain of dependent steps, counting both precedence and reliance on a common resource. That is the gap Kelley and Walker flagged in 1959, when they wrote that manpower and equipment were missing from their method. The full story is in the history of Critical Chain Project Management. Where computer scheduling went after CPM, including the AI schedulers of the 1980s, is traced in the history of AI in project management.

For a view across the whole arc, see the evolution of project management methodologies.

What this means for teams today: keep the baseline, the updates and the reasons for changes. When a schedule is the only dated record of what you meant to do, it will be read by people who weren't there.

Key People

Key Facts

• Kelley and Walker's paper records that Remington Rand and duPont began a cooperative effort in May 1957 on duPont's UNIVAC I.
• At Louisville, average shutdown time fell from 125 hours to 93, with 78 hours projected.
• The first live test network had 393 jobs, 156 restraints and 297 dummies, and only seven of 156 deliveries were critical.
• Mauchly Associates had trained close to 1,000 people by the end of 1962, according to Walker.
• In a 1974 survey, over 80 percent of 235 respondents among the 400 largest U.S. construction firms used critical path methods.
• The SCL Protocol says contractor programmes should in most cases be based on CPM.

The Pattern Behind the History

This is our synthesis, not published research. Five points repeat in the history of CPM:

• A narrow pain starts it. A chemical company wanted shorter shutdowns and cheaper construction, not a new discipline.
• Math first, picture second. Kelley built a linear program and drew the arrow diagram to explain it to managers.
• Twin inventions happen. CPM and PERT started independently from the same need, and Fondahl and Roy reached node-based networks in the same year, 1958.
• The simple core survives. Network, early and late times, and float are what stayed. Mauchly Associates shifted the focus to schedule over cost, and PERT, Weaver says, has died out.
• The notation follows the software. Precedence diagramming won because IBM coded it, and the method's later life is in contracts and claims.

What It Teaches Teams Today

• Draw the network before you set dates. Kelley and Walker split planning from scheduling for a reason. First decide what must precede what, then add durations.
• Don't crash everything. DuPont found only about 10 percent of jobs were critical. Spend extra effort on those, and treat the rest as float.
• Check what the method leaves out. The 1959 paper says people and equipment were absent. Add resources before you trust a date. Mapping task dependencies and reading float and slack help, but they don't allocate people.
• Keep the record. A baseline plus dated updates is both a management tool and, if the project is disputed, evidence.
• Match the tool to your uncertainty. For work with stable logic, CPM is still the right first move. For work full of discovery, it is a poor fit.