Despite its popularity in clinical research, the emotional Stroop task's reliability in patient groups is unknown. Given the low reliability of interference scores in healthy subjects, correlations with other variables pose a problem, especially as reliability in clinical samples is unknown. To assess reliability in panic disorder for the first time, we used the spilt-half method in two independent samples of patients and controls. As expected, only patients showed the behavioral interference effect. Reliability of interference scores (i.e. mean response latency emotional minus neutral words) was insufficiently low for patient and control samples; however, reliability scores derived from the conditions' response latencies (i.e. mean response latency emotional or neutral words) were much higher. The assumption that reliability scores in patients might differ from controls was not supported. This finding questions the use of correlations with external variables and suggests the use of response latencies instead of interference scores.
