The assessment of physical activity at a single time to evaluate its association with cancer may be limited, since such a measure may not adequately reflect activity over the long term. To overcome this limitation, we studied 17,148 Harvard alumni aged 30-79 years who were followed prospectively for the occurrence of colon cancer (n = 225) and rectal cancer (n = 44) from 1965 through 1988. Physical activity, based on self-reported stair climbing, walking, and sports play, was assessed in either 1962 or 1966 (1962/1966) and again in 1977. The increased activity evaluated using either assessment (1962/1966 or 1977) taken alone was not associated with risk of colon cancer. However, alumni who were highly active (energy expenditure of greater than 2500 kilocalories/wk) at both assessments had half the risk of developing colon cancer relative to those who were inactive (less than 1000 kilocalories/wk) at both assessments (age-adjusted rate ratio = 0.50; 90% confidence interval = 0.27-0.93), whereas those who were moderately active (1000-2500 kilocalories/wk) at both assessments had an age-adjusted rate ratio of 0.52 (90% confidence interval = 0.28-0.94). We conclude that either consistently higher levels of activity are necessary to protect against colon cancer or combining two assessments increased the precision of physical activity measurement. We found no evidence that increased physical activity protected against rectal cancer.