Previous meta-analyses have shown the effectiveness of cognitive behavioral therapy for insomnia (CBT-I). However, conclusive information about therapeutic effects (especially during follow-up), effect sizes of objective sleep parameters and self-rating scales, and the problem of publication bias has not been obtained. We conducted a meta-analysis focusing on these issues. We identified 14 randomized controlled studies published between 1990 and 2009 that fulfilled our selection criteria. Intra-group comparison of CBT-I and comparison between CBT-I and control groups were performed on these studies. The intra-group comparison revealed that the effect sizes of CBT-I for subjective sleep variables from sleep diaries were medium to large at the end point of treatment, and these effect sizes were favorably maintained on follow-up. A between-group comparison revealed that CBT-I was more effective than the control for subjective sleep variables at the end of treatment and that its effectiveness was also recognized on follow-up. With regard to self-rating scales, as compared to the control group, the effect sizes in the CBT-I group were medium to large both at the end of treatment and on follow-up. However, there were problems of publication bias in some of the subjective or objective sleep variables. The abovementioned results support the effectiveness of CBT-I for the treatment and prevention of relapse of primary insomnia despite the existence of a certain publication bias.