The present study, which is part of the research of the IMPETUS (“An Integrated Approach to the Efficient Management of Scarce Water Resources in West-Africa") project, shows an analysis of wet and dry periods in north-western Africa both from observed rainfall and from scenarios obtained from an ensemble study of the regional climate model REMO. One question is how different sources of data with different quality and different statistical characteristics can be interpreted with respect to the past and future development of rainfall variability. In order to compare results from our different sources (GHCN station rainfall data, CRU gridded rainfall, and model data) a normalization of rainfall data using the standardized precipitation index (SPI) has been undertaken. Three regions of similar rainfall variability are marked by a principal component analysis of rainfall data: The Atlantic, Mediterranean, and South-of-the-Atlas-Mountain regions. For these three regions, observed SPI indices for the period 1901-2007 and return times of dry and wet spells are discussed. Return times have been assessed using, among others techniques, a fit of the generalized Pareto distribution (GPD) to the SPI values above a certain threshold. The REMO climate for the control period 1961-2000 exhibits a bias of the regional model towards dryer conditions as revealed by the Köppen climate classification of the very heterogeneous investigation area. Despite the model dry bias, estimated return values of regional climate model and observational data are in good agreement. Based on six realisations of the IPCC A1B/B1 climate scenarios for the period 2001-2050, the SPI series was evaluated with respect to future return times of sufficiently high/low values. The future scenario data for 2031-2050 clearly showed more frequent occurrences of dry periods in all regions and nearly unchanged wet period occurrences in the Atlantic and South-of-the-Atlas regions. In the Mediterranean region, the wet period frequencies seem reduced in regional climate model scenarios.