This paper presents an application of homogeneous Markov chain model to time series of SPI in order to characterize droughts in Serbia. The long-term probabilities of occurrence of different drought severity classes, recurrence times, expected residence time in each drought class and expected first passage time to reach the non-drought state were computed. Results are presented in a form of maps and the drought-prone areas are identified. Results of the performed analysis are useful in the context of drought early warning systems and in a planning of drought mitigation practices.