This paper presents a non-homogeneous Markov chain approach for analyzing drought characteristics using the Palmer Drought Severity Index (PDSI). The probability mass functions of occurrence of different drought severity classes, their durations, and times of return to a particular drought class are obtained and are in turn utilized to generate the needed statistics and forecasts. Two methods of forecasting drought severity classes for one, two, and three months lead times are put forward. The methodology is applied to ninety six years of PDSI data corresponding to two climatic divisions in Virginia, USA. Comparison between the analytical results and the empirical estimates supports the utility of the method. The method can be used in a planning mode for developing buffer storage in drought prone regions and in an operational mode for optimal rationing of water among competing needs as drought progresses.