Mechanisms of mitochondrial superoxide formation remain poorly understood despite considerable medical interest in oxidative stress. Superoxide is produced from both Complexes I and III of the electron transport chain, and once in its anionic form it is too strongly charged to readily cross the inner mitochondrial membrane. Thus, superoxide production exhibits a distinct membrane sidedness or "topology." In the present work, using measurements of hydrogen peroxide (Amplex red) as well as superoxide (modified Cypridina luciferin analog and aconitase), we demonstrate that Complex I-dependent superoxide is exclusively released into the matrix and that no detectable levels escape from intact mitochondria. This finding fits well with the proposed site of electron leak at Complex I, namely the iron-sulfur clusters of the (matrix-protruding) hydrophilic arm. Our data on Complex III show direct extramitochondrial release of superoxide, but measurements of hydrogen peroxide production revealed that this could only account for approximately 50% of the total electron leak even in mitochondria lacking CuZn-superoxide dismutase. We posit that the remaining approximately 50% of the electron leak must be due to superoxide released to the matrix. Measurements of (mitochondrial matrix) aconitase inhibition, performed in the presence of exogenous superoxide dismutase and catalase, confirmed this hypothesis. Our data indicate that Complex III can release superoxide to both sides of the inner mitochondrial membrane. The locus of superoxide production in Complex III, the ubiquinol oxidation site, is situated immediately next to the intermembrane space. This explains extramitochondrial release of superoxide but raises the question of how superoxide could reach the matrix. We discuss two models explaining this result.