The accident occurred in the TMI-2 reactor (the plant had two reactors; TMI-1 was down for refueling at the time) and had a number of primary causes, related both to technical malfunction and human error. It began when the plant's main feedwater pumps in the secondary non-nuclear cooling system failed at about 4:00 a.m. on March 28, 1979. This failure was due to either a mechanical or electrical failure in the condensation system and caused a reduction in feedwater flow which prevented the steam generators from removing heat. The auxiliary (backup) feedwater system had been inadvertently left valved-out after an earlier maintenance activity. First the turbine, then the nuclear reactor automatically shut down. Immediately, the pressure in the primary system (the nuclear portion of the plant) began to increase. In order to prevent that pressure from becoming excessive, the pressurizer relief valve (a valve located at the top of the pressurizer) opened. The valve should have re-closed when the pressure decreased by a small amount, but it did not. The only signals available to the operators showed the valve as being closed, but in fact only the signal to close the valve was sent; the system did not check that the valve was actually closed. The "positive feedback" lamp in the control room indicating the true position of the valve (a Pressure Operated Relief Valve, or PORV) was eliminated in original construction to save time and has been backfitted to all other similar plants. As a result of this design error, the valve remained open and caused the pressure to continue to decrease in the system.

It should be noted that the operators and emergency operating procedures (EOPs) did not recognize the accident as a classic LOCA (Loss of Coolant Accident) since they had no dependable instrumentation to indicate the loss of primary water or non-ambiguous reactor level indication (see Aftermath).

Meanwhile, another problem appeared elsewhere in the plant with the emergency feedwater system (the backup to the main feedwater). It had been tested 42 hours prior to the accident. As part of the test, a valve is closed and then reopened at the end of the test. But this time, through either an administrative or human error, the valve was not reopened. This prevented the emergency feedwater system from functioning during the accident. The valve was discovered closed about eight minutes into the accident. Once it was reopened, the emergency feedwater system began to work correctly, allowing cooling water to flow into the steam generators.

As the system pressure in the primary system continued to decrease, voids (areas where there is no water present) began to form in portions of the system other than the pressurizer. Because of these voids, the water in the system was redistributed and the pressurizer still had water available. The turbulence of this water blew out the stuck-open valve and caused the level indicator to think the pressurizer was full. Thus the level indicator, which tells the operator the amount of coolant capable of heat removal, incorrectly indicated the system was full of water. Therefore, the operator stopped adding water — by turning off the Emergency Core Cooling pumps, which had automatically come on. He was unaware that, because of the stuck valve, the indicator could, and in this instance did, provide false readings.

After almost eighty minutes of slow temperature rise, the primary loop pumps began to cavitate as steam, rather than water, began to pass through them. The pumps were shut down, and it was believed that natural circulation would continue the water movement. Steam in the system locked the primary loop, and as the water stopped circulating it was converted to steam in increasing amounts. About 130 minutes after the first malfunction, the top of the reactor core was exposed and the heat and steam drove a reaction involving hydrogen and radioactive gases with the zirconium nuclear control rod cladding. The quench tank (collecting the discharge from the PORV) overfilled, its relief diaphragm ruptured, and radioactive coolant began to leak out into the general containment building. At 6 a.m. there was a shift change in the control room. A new arrival noticed that the temperature in the holding tanks was excessive and used a backup valve to shut off the coolant venting, but around 950 m³ (250,000 US gallons) of coolant had already leaked from the primary loop. It was not until 165 minutes after the start of the problem that radiation alarms activated as contaminated water reached detectors — by that time the radiation levels in the primary coolant water were around 300 times expected levels, and the plant was seriously contaminated.

At 7:00 AM a "Site Area Emergency" was declared. At 7:24 AM the incident was upgraded to a "General Emergency". Harrisburg radio station WKBO announced a problem with the plant at 8:25 AM. The Associated Press announced the general emergency at 9:00 AM.

It was still not clear to the control room that the primary loop water levels were low and that over half of the core was exposed (a LOCA). A group of workers took manual readings from the thermocouples and obtained a sample of primary loop water. Around seven hours into the emergency, new water was pumped into the primary loop. The backup relief valve was opened to reduce pressure. At around nine hours the hydrogen within the reactor building ignited and burned, but this was largely unnoticed. After almost sixteen hours the primary loop pumps were turned back on and the core temperature began to fall. A large part of the core had melted, and the system was still dangerously radioactive. Over the next week the steam and hydrogen were removed from the reactor using a recombiner and, more controversially, by venting straight to the atmosphere. It is estimated that a maximum of 13 million curies (480 petabecquerels) of radioactive noble gases were released by the event, though very little of the hazardous iodine-131 was released.

The molten fuel did not break through in a "China Syndrome". "Despite melting of about one-third of the fuel, the reactor vessel itself maintained its integrity and contained the damaged fuel."