In this study, a highly flexible form of nonlinear time series models called artificial neural networks (ANNs) are employed to predict fluctuations in economic activity in selected members (Armenia, Azerbaijan, Georgia, Kazakhstan, and Kyrgyzstan) of the Commonwealth of Independent States (CIS) using macroeconomic time series [treasury bill rate (T-bill), long term bond rate (BondLT), money supply (MS), industrial production (IP), spread (10-year treasury bond rate less 3-month treasury bill rate), BRTB (bank rate less 3-month treasury bill rate), and GDP growth rate]. Forecasting recessions being very important though challenging, recessions in the selected countries are modeled recursively 1---10 quarters ahead out-of-sample using ANNs in conjunction with macroeconomic time series for all the countries. The out-of-sample forecast results show that in general no single macroeconomic variable employed appears to be useful for predicting recessions in any of the series. However, for Armenia, the treasury bill rate, industrial production, money supply, and the spread (the yield curve) are candidate variables for predicting recessions 1---10 quarters ahead. For Georgia, Kazakhstan, and Kyrgyzstan, the treasury bill rate and money supply series are candidate variables for predicting recessions 1---10 quarters ahead.