Background. The demand for time-consuming psychotherapy of phobia/panic exceeds the supply of trained therapists. Delegating routine therapy aspects to a computer might ease this problem.Method. Ninety-three out-patients with phobia or panic disorder were randomized in a 2[ratio ]2[ratio ]1 ratio to have self-exposure therapy guided either mainly by a stand-alone computer system (FearFighter) or entirely face-to-face by a clinician, or to have mainly computer-guided self-relaxation as a placebo. Both computer groups (FearFighter and relaxation) had brief back-up advice from a clinician. Primary outcome measures were self- and blind-assessor ratings of Main Problem and Goals, and Global Phobia.Results. Drop-outs occurred significantly more often in the two self-exposure groups (43% if mainly computer-guided, 24% if entirely clinician-guided) than with self-relaxation (6%); the difference between the two self-exposure groups was not significant. Even with all drop-outs included, the mainly computer-guided exposure group and the relaxation group had 73% less clinician time per patient than did the entirely clinician-guided exposure group. The two self-exposure groups had comparable improvement and satisfaction at post-treatment and at 1-month follow-up, while relaxation was ineffective. Mean improvement on the primary outcome measures (self- and assessor-rated) was 46% computer, 49% clinician, 9% relaxation at post-treatment (week 10) and 58% computer, 53% clinician and −4% relaxation at 1-month follow-up (week 14). Mean effect sizes on the primary outcome measures were 2·9 computer, 3·5 clinician and 0·5 relaxation at post-treatment; and 3·7 computer, 3·5 clinician and 0·5 relaxation at 1-month follow-up. The assessor did not rate patients at follow-up.Conclusions. Despite its (non-significantly) higher dropout rate, self-exposure therapy for panic/phobia cut clinician time per patient by 73% without losing efficacy when guided mainly by a computer rather than entirely by a clinician. The finding needs confirmation at a follow-up that is longer and includes a blind assessor. Self-relaxation had the highest rate of completers but was ineffective.