Many patients seeking primary care for low back pain continue to report disability several years after their initial visit. The aims of this study were to assess the independent predictive value of a number of potential predictive factors for disability at the 1-year and 5-year follow-ups, and to examine whether prediction models were improved by replacing baseline health-state-related variables with corresponding variables after treatment. A further aim was to describe possible differences between those on sick leave, early retirement or disability pension, and those who were not. Baseline factors were age, gender, self-reported physical-activity-related and work-related factors, expectations of treatment, similar problems previously, duration of episode, more than one localization, sick leave, pain frequency, disability, and well-being. The study sample comprised 148 participants in a previous randomized trial who were eligible for sick-leave benefits. Multiple logistic regression was used to identify predictive factors. At the 5-year follow-up, 37% (n = 19/52) of the patients with disability were on sick leave or were receiving early retirement or disability pension. For those without disability the corresponding figure was 9% (n = 8/92). Being a woman, duration of the current episode, similar problems during the previous 5 years, exercise level before the current episode, pain frequency at baseline, and disability after treatment emerged as predictive factors for disability at the 5-year follow-up. Replacing baseline health-state-related measures with corresponding measures after the treatment period, and adding physical-activity-related and possibly work-related factors might improve the likelihood of predicting future disability.