This paper extends the Blinder-Oaxaca decomposition technique to logit and probit models, allowing for the analysis of racial and gender gaps in binary outcomes. The original technique is used to decompose differences in outcomes into contributions from group differences in characteristics and differences in the effects of those characteristics. However, it cannot be directly applied to binary outcomes with coefficients from logit or probit models. The author proposes a method that uses estimates from logit or probit models to perform the decomposition. The decomposition technique is applied to analyze racial differences in computer ownership. The results show that income and education are the main factors contributing to the racial gap in computer ownership. The decomposition also reveals that group differences in family characteristics explain a significant portion of the gap, while differences in regional distributions have a smaller impact. The paper compares the results of the non-linear decomposition technique with those from a standard Blinder-Oaxaca decomposition using a linear probability model. The results are similar, but the non-linear technique provides more accurate estimates, especially when the racial gap is located in the tails of the distribution or when racial differences in independent variables are large. The paper also discusses the effect of variable ordering on the decomposition results. The results are not substantially different when the order of variables is reversed, but the differences are worth noting. The author suggests experimenting with different variable orders to confirm the robustness of the results. Finally, the paper addresses the use of sample weights in the decomposition technique. The results are not substantially different when sample weights are used, but the interpretation of the decomposition changes slightly. The author finds that estimates using both white and black sample weights from the CPS do not differ substantially from unweighted estimates.This paper extends the Blinder-Oaxaca decomposition technique to logit and probit models, allowing for the analysis of racial and gender gaps in binary outcomes. The original technique is used to decompose differences in outcomes into contributions from group differences in characteristics and differences in the effects of those characteristics. However, it cannot be directly applied to binary outcomes with coefficients from logit or probit models. The author proposes a method that uses estimates from logit or probit models to perform the decomposition. The decomposition technique is applied to analyze racial differences in computer ownership. The results show that income and education are the main factors contributing to the racial gap in computer ownership. The decomposition also reveals that group differences in family characteristics explain a significant portion of the gap, while differences in regional distributions have a smaller impact. The paper compares the results of the non-linear decomposition technique with those from a standard Blinder-Oaxaca decomposition using a linear probability model. The results are similar, but the non-linear technique provides more accurate estimates, especially when the racial gap is located in the tails of the distribution or when racial differences in independent variables are large. The paper also discusses the effect of variable ordering on the decomposition results. The results are not substantially different when the order of variables is reversed, but the differences are worth noting. The author suggests experimenting with different variable orders to confirm the robustness of the results. Finally, the paper addresses the use of sample weights in the decomposition technique. The results are not substantially different when sample weights are used, but the interpretation of the decomposition changes slightly. The author finds that estimates using both white and black sample weights from the CPS do not differ substantially from unweighted estimates.