This paper contributes to the empirical literature on Islamic finance by doing comparison of Islamic and conventional banks in Pakistan over the period 2005-2014. We apply both non-parametric and parametric classification methods (neural network, linear discriminant analysis, and logistic regression) to investigate whether financial ratios can be used to distinguish between Islamic and conventional banks. Univariate analysis reveals that Islamic banks are less profitable, better capitalized, more liquid, and have low level of credit risk as compared to their conventional counterparts. We also find that Islamic banks have more operating leverage in comparison to conventional banks. The results from classification techniques show that the two types of banks may be distinguished in terms of insolvency risk, credit risk, efficiency, and operating leverage, but not in terms of liquidity and profitability. More interestingly, we find that the financial crisis has a negative effect on the profitability of both Islamic and conventional banks. Lastly, the results show that the neural model obtained higher classification accuracies as compared to other models used in the study.