The reliability of long-term recall of physical activity participation was examined in 322 women and 129 men in a worksite health study conducted at the Liberty Corporation, Greenville, South Carolina during 1976-1987. Leisure time physical activity was assessed at baseline; and energy expenditure in total, light, moderate, and vigorous activities was calculated. The long-term recall of baseline activities was determined 1-10 years after the examination. The relation between actual baseline and recalled activity was positive and in most cases the coefficients were statistically significant at p less than 0.05. The correlations were modest, most in the range of 0.20 to 0.50. Percent agreement between baseline and recalled activity generally ranged from 60 to 75%. Multiple regression analyses suggested that recalled activity was a significant predictor of baseline activity, but recall interval and age were not important contributors to the regression model. R2 values for the model were 0.10 for light activity and 0.26 for vigorous activity. Questionnaire assessment of long-term physical activity recall appeared to be reliable, length of recall interval up to 10 years was not an important factor, and recall of vigorous activity was more accurate than for less intensive activities.