This study presents the methodology of sensitivity analysis and explores whether it can be an alternative evaluation criterion as well as a tool to ”read” artificial neural networks' knowledge. The simulation of the Black-Scholes formula is employed for this object. Since, in the Black-Scholes formula, the mapping relationship between the call price and five relevant variables is a mathematically close form, it is feasible to verify the validity of the methodology of sensitivity analysis. The experiment results are promising; they show that both values of the sensitivity analysis and the partial derivative of the Black-Scholes formula are consistent. Furthermore, the sensitivity analysis can be an alternative criterion for comparing the effectiveness of ANNs.