The synthesis of ATP from highly enriched [18O]Pi by submitochondrial particles driven by succinate oxidation produces distributions of 18O-labeled ATP species that deviate from the distributions predicted by a simple model for the exchange. Control experiments indicate no change in isotopic distribution when [18O]ATP is synthesized from [18O]ADP by adenylate kinase, which is bound to the submitochondrial particles. The observed deviations are in the opposite direction from that produced by heterogeneity due to multiple pathways for ATP synthesis. Two types of complex models can account for the observed deviations. One model has nonequivalence of the Pi oxygens during the exchange reaction, due to incomplete randomization of the Pi oxygens during the reversible cycles of hydrolysis and synthesis of bound ATP. The other model assumes that, during each turnover, a slow transition must occur between a high-exchange and a low-exchange pathway.