We performed a retrospective study of 616 patients (636 shoulders) who had subacromial impingement syndrome to assess the results of non-operative treatment. The diagnosis was made on the basis of a positive impingement sign and the absence of other abnormalities of the shoulder, such as full-thickness tears of the rotator cuff, osteoarthrosis of the acromioclavicular joint, instability of the glenohumeral joint, or adhesive capsulitis. All patients were managed with anti-inflammatory medication and a specific, supervised physical-therapy regimen consisting of isotonic exercises for strengthening of the rotator cuff. The average duration of follow-up was twenty-seven months (range, six to eighty-one months). Over-all, 413 patients (67 per cent) had a satisfactory result. One hundred and seventy-two patients (28 per cent) had no improvement and went on to have an arthroscopic subacromial decompression. Thirty-one patients (5 per cent) had an unsatisfactory result but declined additional treatment. Seventy-four (18 per cent) of the 413 patients who had a successful result had a recurrence of the symptoms during the follow-up period; the symptoms resolved with rest or after resumption of the exercise program. The patients were stratified according to age, the duration of symptoms, and acromial morphology. Patients who were twenty years old or less and those who were forty-one to sixty years old fared better than those who were twenty-one to forty years old. Patients who were more than sixty years old had the poorest results. Sixty-seven (78 per cent) of the eighty-six patients in whom the symptoms had been present for less than four weeks had a satisfactory result, compared with 144 (63 per cent) of the 228 who had had the symptoms for one to six months and with 202 (67 per cent) of the 302 who had had the symptoms for more than six months. Thirty-two (91 per cent) of the thirty-five patients who had a type-I acromion had a successful result, compared with 173 (68 per cent) of the 256 who had a type-II acromion and with 208 (64 per cent) of the 325 who had a type-III acromion. Shoulder dominance, gender, and concomitant tenderness of the acromioclavicular joint did not affect the result significantly (p = 0.084, 0.555, and 0.365, respectively).