The Cumulative Sum Chart for process control is designed to detect relatively small shifts in the mean of the process variable. In this paper, we show that by using samples of two different sizes, rather than a fixed sample, improvements in the chart's ability to detect small shifts are possible. These improvements are obtained without increasing the likelihood of a false alarm when the process mean is on target. The average run length values for the adaptive sample scheme are compared to those for a fixed sample scheme for different magnitudes of shifts in the process mean. A fast initial response scheme is presented. An example from a manufacturing process is used to illustrate the advantages of the adaptive sampling scheme.