Conflicting evidence exists about the effectiveness of spinal manipulation. To validate a manipulation clinical prediction rule. Multicenter randomized, controlled trial. Physical therapy clinics. 131 consecutive patients with low back pain, 18 to 60 years of age, who were referred to physical therapy. Patients were randomly assigned to receive manipulation plus exercise or exercise alone by a physical therapist for 4 weeks. Patients were examined according to the clinical prediction rule criteria (symptom duration, symptom location, fear-avoidance beliefs, lumbar mobility, and hip rotation range of motion). Disability and pain at 1 and 4 weeks and 6 months were assessed. Outcome from spinal manipulation depends on a patient's status on the prediction rule. Treatment effects are greatest for the subgroup of patients who were positive on the rule (at least 4 of 5 criteria met); health care utilization among this subgroup was decreased at 6 months. Compared with patients who were negative on the rule and received exercise, the odds of a successful outcome among patients who were positive on the rule and received manipulation were 60.8 (95% CI, 5.2 to 704.7). The odds were 2.4 (CI, 0.83 to 6.9) among patients who were negative on the rule and received manipulation and 1.0 (CI, 0.28 to 3.6) among patients who were positive on the rule and received exercise. A patient who was positive on the rule and received manipulation has a 92% chance of a successful outcome, with an associated number needed to treat for benefit at 4 weeks of 1.9 (CI, 1.4 to 3.5). The response rate for the 6-month follow-up resulted in inadequate power to detect statistically significant differences for some comparisons. The spinal manipulation clinical prediction rule can be used to improve decision making for patients with low back pain.