OBJECTIVES: Phototherapy is widely used for the treatment of seasonal affective disorder (SAD). Evidence for its efficacy comes primarily from crossover trials in which patients received phototherapy or a placebo treatment, and then crossed over to receive the opposite treatment. METHODS: Our meta-analysis of controlled trials of phototherapy found statistically significant heterogeneity among study results (Q = 20.39, p = 0.08). Study design (parallel or crossover) and blinding both influenced the magnitude of the reported effect. Unblinded, crossover studies had significantly larger treatment effects than blinded parallel studies (p = 0.01). Three crossover studies reported data separately for treatments before and after patients crossed over. In these studies, treatment effects prior to the crossover were similar to treatment effects observed in parallel trials (p = 0.36). However, treatment effects from the second, post-crossover period are significantly greater than in parallel trials (p = 0.001). RESULTS: Further examination of these trials showed that not only were differences between treated and placebo patients significantly greater after the crossover than before (p = 0.04), but depression ratings after active treatment were significantly lower among patients receiving active treatment after crossover than among those receiving active treatment before crossover (p = 0.02). These differences persist when one examines number of responders to therapy rather than raw depression ratings (p = 0.03 or p = 0.002, depending on the definition of "responder" used. Moreover, placebo responses are larger when patients receive placebo treatment before crossover compared to after crossover, although this difference is statistically significant only for one definition of "responder" (p = 0.04). CONCLUSION: These ordering effects may be the result of loss of blinding after crossover. These results suggest that proper patient blinding is essential in the evaluation of interventions for affective or other psychological disorders. Crossover studies, which because patients are exposed to both treatments cannot be blinded, are inappropriate for investigating the efficacy of such treatments.