This was a retrospective study of 115 patients who underwent 124 cycles of ICSI using surgically retrieved spermatozoa. The objective was to compare the results of ICSI in patients with obstructive azoospermia using epididymal spermatozoa (36 cycles) or testicular spermatozoa (58 cycles) with ICSI in patients with non-obstructive azoospermia using testicular spermatozoa (30 cycles). When epididymal spermatozoa were used for ICSI, the fertilization rate per injected metaphase-II oocyte and the clinical pregnancy rate per ICSI cycle were 60.4 and 25%, respectively. When testicular spermatozoa were used in obstructive cases, the fertilization rate and pregnancy rate were 57.9 and 34.5%. In non-obstructive cases the fertilization and pregnancy rates were 41.2 and 16.6%. When patients with obstructive azoospermia were regrouped according to the cause of obstruction, the fertilization and pregnancy rates were 59.1 and 35.1% in acquired obstruction and 58.7 and 24.3% in congenital obstruction. The fertilization and pregnancy rates were not statistically different (p > 0.05) when testicular or epididymal spermatozoa were used in obstructive cases; neither was statistically different (p > 0.05) when compared in patients with congenital and acquired obstruction. On the other hand, the fertilization and pregnancy rates in cases with non-obstructive azoospermia were significantly lower (p < 0.05) than in obstructive cases.