This paper proposes the reliability-based optimization design model that has no problem of convergence compared to the traditional one for the rock-socketed pile which has fewer design variables and more random variables controlling load effect and resistance, analyzes a case about the reliability-based optimization design with the model calculated by matlab, and studies the sensitivity of random variables affecting the design variables. It is concluded that the target reliability index can’t be met by adjusting the design parameters, resulting from that the statistical and mechanical parameters of the load and rock are sensitive. It is suggested to strengthen the statistical work of the parameters.