The problem of monitoring autocorrelated process data is considered. A Markov chain approach is used to determine the performance of several forecast-based monitoring schemes for various types of process shifts and evaluation criteria. The performance of the monitoring schemes is shown to depend on the process model, type of shift, and evaluation criteria. The combined exponentially weighted moving average-Shewhart (CES) control chart applied to appropriate one-step-ahead forecast errors is recommended for general use, and a design procedure is provided.