Hollerith, Punched Cards and the 1890 Census

By Victor Hoang · Updated Oct 22, 2026

The 1880 census took years to tabulate by hand. Anyone could see the next count would be bigger. The Computer History Museum notes that estimates warned the 1890 census would not be finished before the 1900 census began.

Herman Hollerith's punched cards were first used at scale in the 1890 US Census, and his company became IBM in 1924. The card is older: weavers in Lyon steered looms with perforated paper in the 1720s. Hollerith turned the hole into a unit of data about a person. Columbia's history puts the 1880 census at about 50 million people, seven years of tabulation and around $5.8 million. For the whole arc of the field, see The History of Computers and the computer timeline.

Timeline at a Glance

The Loom Years (1725-1834)

Lyon weavers invent the hole as an instruction

The story starts in a silk workshop, not a government office. On a Lyon drawloom, an operator used to lift the warp threads by hand using cords. According to Grace's Guide, Basile Bouchon, an organ maker's son working in Lyon's silk trade, used a punched paper tape to select those threads in 1725, partly automating the setup. His assistant Falcon arranged the holes in rows on rectangular cards linked into an endless loop in 1728, which controlled more cords and cut lifting errors. The same page notes that Vaucanson made an attempt at automating the process in 1745, and that the Jacquard mechanism was in production by 1805.

The turning point of the era is Jacquard. The Computer History Museum dates his loom to 1801 and calls it an early example of automatic control and of stored data in commercial production, while Grace's Guide has it in production by 1805, so the date depends on whether you count the demonstration or the factory. Nobody called any of this computing. The holes carried a pattern of flowers, not a population.

Then a mathematician noticed. Charles Babbage designed an engine in the 1830s that took its instructions from cards, and the Museum records that the Analytical Engine, never built, included a card reader. That story, with the operation cards and variable cards, is in Charles Babbage and the Analytical Engine and in Ada Lovelace and the first computer program. Babbage's design was paper only, and the cards stayed in the textile mills.

What this era solved was storing a procedure outside the operator's head. What it left open was a different use of the card: recording facts rather than instructions.

For teams today: the best idea in a process is often borrowed from another trade. When you hit a stubborn problem, look at how an unrelated industry has already solved the same shape of problem.

The Census Problem (1880-1889)

A young statistician and a patent filing in Washington

The turning point is a job. Columbia's account says Hollerith graduated at 19, and his engineering professor William P. Trowbridge offered him a job as a statistician on the 1880 census. The IEEE Computer Society pioneer page lists him at the Census Bureau from 1879 to 1882, working on statistical analysis under John Shaw Billings. It also says he taught at MIT in 1882 and 1883 and worked as a Patent Office examiner from 1883 to 1886.

What he saw was an organization drowning in manual counting. The MIT Technology Review says Billings, who headed the census office's Division of Vital Statistics, suggested storing information as notches in cards. The same article tells the often repeated story that Hollerith watched a train conductor punch tickets to record the rider, and notes that the timing of that observation is unclear. IBM's history repeats the conductor story as inspiration. Treat it as the family legend. Billings's notched-card suggestion and the conductor's ticket are the two origins the sources offer, and neither source ranks them.

The legal record is clearer. The Technology Review says he was an assistant examiner at the Patent Office from May 1883 until he resigned on 31 March 1884, and filed his first card application in September 1884. Google Patents lists US 395,781, "Art of Compiling Statistics," as granted on 8 January 1889, with a filing date of 8 June 1887. That is the filed-versus-granted distinction at work: the idea dates from 1884, the application on that patent record from 1887, and the grant from 1889, when three patents were issued the same day. The patent text describes cards punched with one hole per fact, read by a press with spring-loaded pins dipping into mercury cups. It gives the example of Baltimore mortality records, and the ETHW account describes a practical test of the equipment for the Baltimore health department in 1886.

Then the contest. The Technology Review says that in 1889 the census office held a competition for the machines that would tabulate the eleventh census, and Hollerith's system won. He kept his business model simple: he rented his machines with service and support included and required buyers to purchase his own high-quality cards, after the census office's inferior paper cards left fibers in the mercury.

This era turned an observation into a patented method. The new problem was whether it would hold up under the weight of a whole nation.

For teams today: the inventor who wins is often the one who worked inside the problem for years first. Spend a season inside the process you want to fix before you design the fix.

The Count of 1890 (1890-1895)

Sixty million cards and a pantograph

The turning point is the census itself. The Computer History Museum says the count had grown from under four million in 1790 to 63 million by 1890. Clerks recorded each person on a schedule, then operators punched a card using a pantograph. The Museum describes the pantograph as a device in which the operator aligned a pin with a hole in a guide plate and pushed hard to punch a round hole in a stiff card. It adds that the early punch was slow compared to later keyboard-driven punches. The card was roughly the size of a dollar bill of the time, so existing currency drawers and bins could hold it.

The tabulator itself did the rest. Pins passed through the holes into cups of mercury, completing an electrical circuit that advanced a dial. The Museum puts the work at about 60 million cards punched and operators handling about 7,000 cards a day, at least ten times faster than manual methods. Columbia counts 63 million cards on 43 tabulating machines, each card 6 inches wide and 3 1/4 inches high. The two institutions differ on the card total by a few million, so the safe description is "roughly 60 to 63 million."

How long it took depends on whom you ask. The Library of Congress says the system reduced tabulation to about 2 1/2 years, down from eight years for 1880. Columbia says the 1890 count took under three years, against seven for 1880. IBM's own history says only that the 1890 census finished months ahead of schedule and under budget. The IEEE page says tabulation fell to one-third of the time of 1880. All of them agree the direction was a large drop.

The count also did something the earlier card never had. It sorted people. The Library of Congress notes that a celluloid panel coded for the 1890 census was used with the pantograph, and that Congress used 1890 census data in the 1920s to set immigration quotas. A hole could now answer a question about a human being, and policy could act on the answer.

What this era solved was volume. What it created was a customer who could not be served by a one-off machine: governments, then companies.

For teams today: a pilot that works at your scale proves less than a pilot that works at the scale you will face. Hollerith's machines were tested on Baltimore's deaths, but the census was the test that counted.

The Company Years (1896-1927)

From a rented machine to a holding company called C-T-R

The turning point is the year the invention needed customers other than the government. Columbia says Hollerith founded the Tabulating Machine Company in 1896, while the Computer History Museum dates the founding to 1895, and the ETHW account puts incorporation in 1897. Treat it as 1895 to 1897 depending on the definition. The ETHW account adds that in September 1896 Hollerith signed a contract to audit up to four million waybills for the New York Central Railroad. Cards had left the census.

They also went overseas. The Library of Congress describes the 1903 Philippine census, for which about seven million cards were punched and shipped to Washington, with Hollerith personally modifying machines with new counters and relays. The same post reports that the categories, including race and "civilized" versus "wild," compressed a complicated country into yes-or-no answers. A card records only what someone chose to ask.

Competition and consolidation followed. In 1902 the Census Bureau became permanent, and by 1910 it used Bureau-designed equipment plus rebuilt Hollerith sorters. In 1911 James Powers left the Census Bureau to found a rival firm, the Powers Accounting Machine Company, per the same account. The same year, on 16 June 1911, the Computing-Tabulating-Recording Company was incorporated as a holding company over the Tabulating Machine Company, the International Time Recording Company and the Computing Scale Company. The Computer History Museum records that Hollerith received $1.2 million for his shares and left the company soon after Tom Watson arrived.

Watson came in 1914, when Charles Flint hired him as general manager of a company of about 1,300 employees. IBM's page says C-T-R's revenue doubled within four years of his arrival, and that the company was renamed International Business Machines in 1924. The Technology Review puts the rename "eight years" after 1914, which would be 1922, and IBM's own page gives 1924, so use the latter.

What this era solved was making a government tool into a business. It left a technical limit: the 45-column card with round holes could not hold much.

For teams today: the rented, serviced product with required consumables was Hollerith's best business idea, and it is why the card outlived the original machine. If you sell a tool, ask what the customer must keep buying.

The Eighty-Column Age (1928-1960s)

IBM standardizes the card and the card runs the office

The turning point is 1928. IBM's history says Watson asked two inventors, Clair D. Lake and J. Royden Pierce, to work independently on a card that could hold more. James W. Bryce chose Lake's rectangular-hole design because it needed the fewest changes to existing tabulating machines and worked only with IBM machines. The "IBM Computer Card" arrived in 1928 with 80 columns and 10 rows, and a modified version in 1930 had 12 rows. Before that, cards had evolved from 22 columns and eight punch positions around 1900 to 45 columns of round holes by the late 1920s.

Competitors kept their own formats. The Museum notes that Powers-Samas in the UK kept round holes and made cards with 21, 26, 40, 65 and 80 columns. IBM usually leased equipment for a monthly fee that included maintenance.

For decades the card ran the back office. IBM reports that as late as the mid-1950s, punched card sales made up roughly 20% of its revenue and 30% of its profit. The card became a cultural object too, with its warning "Do not fold, spindle or mutilate." The Museum notes that punching was regarded as clerical work, so most keypunch operators were women. When electronic computers arrived, the card went with them. Each card typically held about one line of code, or about 80 bytes, so programmers fed their work to machines the same way a clerk fed a census schedule. The link is traced in the history of programming languages and generations of computers.

What this era solved was standardization. What it created was dependence: a business whose data lived on cardboard, in tubs, filed in rooms, until storage caught up. The Museum notes that RAMAC, the first disk drive, aimed to eliminate the motorized "tub file" used for punched cards.

For teams today: a de facto standard wins by being compatible with what already works. IBM chose the card design that needed the fewest changes to machines in the field. When you upgrade a format, count how much existing equipment you have to change.

The Long Afterlife (1960s-present)

Magnetic tape replaces the card, but not the habits

The turning point is the replacement by magnetic tape, which IBM says offered faster processing and greater capacity. IBM had stopped selling cards for data processing well before the early 1990s. The deeper story of what came next belongs to the machines described in the history of ENIAC, the history of operating systems and, later, the history of cloud computing.

The card had one more public life. Columbia's account says elections began using Hollerith's tabulation system in the 1960s and it remained the standard for over fifty years, that the 2000 Florida recount popularized the term "hanging chad," and that IBM's Votomatic was last used in two Idaho counties in 2014. The 1890 census method also outlived its own era in government. The Computer History Museum says the Census Bureau still used the same basic system about seventy years after Hollerith's proposal.

His records survive. Columbia reports that his papers are at the Library of Congress and his tabulating machine and framed diploma are at the Smithsonian.

The legacy is not the cardboard. It is the idea that any fact about a person or a thing can be reduced to a code, stored cheaply and counted fast, which is the basis of every database. For a longer view of how that idea turned into computers, see who invented the computer and the history of mechanical calculators.

For teams today: the format dies faster than the habit. People kept thinking in "records" and "fields" long after the cards were gone, so audit the assumptions your data model inherited from its oldest container.

Key People

Key Facts

• Bouchon used punched paper tape on a loom in 1725, and Falcon moved to cards in 1728.
• The 1890 census used roughly 60 to 63 million cards, and operators handled about 7,000 a day.
• Hollerith's US 395,781 was granted on 8 January 1889.
• C-T-R was incorporated on 16 June 1911 and renamed IBM in 1924.
• The 80-column card appeared in 1928.
• In the mid-1950s punched cards were about 20% of IBM's revenue and 30% of its profit.

The Pattern Behind the History

The same four moves repeat in every era above. We call the sequence the Hole-to-Record Cycle. It is our synthesis of the history, not a research finding.

1. Borrow the mechanism. Hollerith took a loom's hole, and a conductor's punch, and turned them to a new job.
2. Prove it on one hard case. Baltimore deaths in 1886, then the whole nation in 1890.
3. Own the supply. Rented machines and required cards made the business recurring.
4. Standardize, then outgrow. The 80-column card won, and later tape and disk replaced it.

What It Teaches Teams Today

• Test at the scale you will face. The census, not the pilot, was the proof.
• Sell the loop, not the box. Hollerith's rented, serviced machines and required cards made the customer relationship last.
• Know which date you mean. Filed, granted, founded, incorporated and renamed were all different years for this story.
• A category is a choice. A card records only the questions someone chose to ask, so decide the fields with care.
• Formats retire, habits stay. Audit what your systems inherited from their oldest container.

Hollerith's card began as a way to count people faster and ended as the template for how organizations store anything. The next time you design a form, a field or a dashboard, remember what the card taught: the question you encode is the answer you will be able to count.