A randomised controlled trial comparing staged teaching of cardiopulmonary resuscitation (CPR) with conventional training provided the additional opportunity to investigate skill acquisition and retention in those attending conventional CPR classes. All subjects were tested immediately after their first instruction period and again at 6-9 months at an unheralded home visit. We were able to assess how far performance was related to poor acquisition of skills and how far it was related to skill decay. Out of 262 subjects who were randomised to receive conventional CPR instruction, 166 were available for home testing at 6-9 months. An invitation to attend for re-training had been accepted by 39 of them. The remaining 127 who attended only a single class comprise the principal study group, with additional comparative observations on the smaller re-trained cohort. Important failings were observed in the acquisition of skills in all modalities tested after the initial instruction. These were particularly marked in skills related to ventilation. Immediately after a class, 68% of trainees performed an effective check of breathing, but only 33% opened the airway as taught and no more than 18% provided an ideal ventilation volume. The technique of chest compression was also less than ideal. Although 80% of subjects placed their hands in an acceptable position, compression to an adequate depth and an adequate rate of compression were achieved by 54 and 63%, respectively. Seventy-eight percent demonstrated a careful approach, and 46% remembered to call for help. A carotid pulse check was simulated by 61% of trainees. When tested 6-9 months later, skill deterioration from this baseline was observed in all modalities tested except for the ventilation volume. The skill decay was significant (P<0.05) for the careful approach, performing an effective breathing check, the carotid pulse check, placing the hands in an acceptable position for chest compression, and compressing at an optimal rate. The minority who attended for re-training showed a trend to protection against skill decay for seven of the ten variables, compared with those who had attended only one training session. This improvement was significant for only two of them, but all were relatively small with limited practical value. Many who attend conventional CPR classes fail to acquire the necessary skills, and the skills that are acquired decline appreciably over the subsequent 6-9 months. The value of conventional re-training was modest in this study of community volunteers.