This article studies the sample sizes of two most commonly used control charts, the X chart and the CUSUM chart. It takes the sampling cost (including the variable and fixed components) into consideration and uses a quadratic loss function as a performance measure. It is interesting to find that if X n = 4 and n CUSUM = 1, the X chart often outperforms the CUSUM chart. Furthermore, the overall performance of both charts can be improved by an optimization design using the quadratic loss function as the objective. The optimal sample size depends on the range of mean shift δ and the ratio between the fixed and variable sampling costs. For general cases (0 < δ ≤ 4), the best sample sizes are X n = 3 for the X chart and n CUSUM = 2 or 3 for the CUSUM chart.Keywords -Control chart, sample size, sampling cost, optimal design, quadratic loss function 978-1-4244-4870-8/09/$26.00 ©2009 IEEE