To identify predictors of back-related long-term functional limitations, 1213 adult enrollees of a Health Maintenance Organization (HMO) in Washington state were interviewed about a month after a consultation for back pain in a primary care setting in 1989-1990, and followed each year thereafter. Out of 100 factors documented at the one-month assessment, measures of somatization, depression, functional limitations, and pain were the strongest predictors of two-year modified Roland-Morris score among a random subsample of 569 subjects. A multiple regression model containing the Symptom Checklist Depression and Somatization scores, the one-month modified Roland-Morris score and the number of pain days in the past six months explained about 30% of the variance in the outcome. Using recursive partitioning, a very simple model was developed to identify patients at high risk of sustaining long-term significant functional limitations. The regression model and the recursive partitioning model were successfully tested in a fresh sample of patients (n = 644). Clinical application of the recursive partitioning model and methodological aspects of this study are discussed.