Randomized clinical trial. Compare outcomes of patients with low back pain receiving treatments matched or unmatched to their subgrouping based on initial clinical presentation. Patients with "nonspecific" low back pain are often viewed as a homogeneous group, equally likely to respond to any particular intervention. Others have proposed methods for subgrouping patients as a means for determining the treatment most likely to benefit patients with particular characteristics. Patients with low back pain of less than 90 days' duration referred to physical therapy were examined before treatment and classified into one of three subgroups based on the type of treatment believed most likely to benefit the patient (manipulation, stabilization exercise, or specific exercise). Patients were randomly assigned to receive manipulation, stabilization exercises, or specific exercise treatment during a 4-week treatment period. Disability was assessed in the short-term (4 weeks) and long-term (1 year) using the Oswestry. Comparisons were made between patients receiving treatment matched to their subgroup, versus those receiving unmatched treatment. A total of 123 patients participated (mean age, 37.7 +/- 10.7 years; 45% female). Patients receiving matched treatments experienced greater short- and long-term reductions in disability than those receiving unmatched treatments. After 4 weeks, the difference favoring the matched treatment group was 6.6 Oswestry points (95% CI, 0.70-12.5), and at long-term follow-up the difference was 8.3 points (95% CI, 2.5-14.1). Compliers-only analysis of long-term outcomes yielded a similar result. Nonspecific low back pain should not be viewed as a homogenous condition. Outcomes can be improved when subgrouping is used to guide treatment decision-making.