The general seasonal Complex Exponential Smoothing (CES) model is presented in this paper. CES is based on conventional exponential smoothing and a theory of complex variables. The proposed seasonal CES can capture known forms of seasonality, as well as new ones that are neither strictly additive nor multiplicative. In contrast to exponential smoothing, CES can capture both stationary and non-stationary processes, giving it greater modelling flexibility. In order to choose between the seasonal and non-seasonal CES a model selection procedure is discussed in the paper. An empirical evaluation of the performance of the model, against ETS and ARIMA, on real data is carried out. The findings suggest that CES simplifies model selection, and as a result the forecasting process, while performing better than the benchmarks in terms of forecasting accuracy.