Aim/purpose – Over estimation and under estimation of the Personal Income Tax (PIT) revenue results in an unstable economy and unreliable statistics in the public domain. This study aims to find a suitable SARIMA and Holt–Winters model that suits the sample monthly data for PIT well enough, from which a forecast can be generated. Design/methodology/approach – This study uses the aspects of time series model (Holt–Winters and SARIMA) and regression models with SARIMA errors to simulate the structure which followed the historical actual realization of PIT. The quarterly data were obtained from quarter1, 2009 to quarter 1, 2017 for the purpose of modelling and forecasting. The data were divided into training (quarter 1, 1995 to quarter 1, 2014) and testing (quarter 2, 2014 to quarter 1, 2017) data sets. The forecast from quarter 2, 2017 to quarter 1, 2020 were also derived and aggregated to annual forecast. Findings – Holt–Winters, SARIMA and Time Series Regression models fitted captured the movement of the historical PIT data with higher precession. Research implications/limitations – The generated forecast is recommended to avoid several model revisions when locating the actual PIT realisation. However, monitoring of this model is crucial as the prediction power deteriorate in a long run. Originality/value/contribution – The study recommends the use of these methods for forecasting future PIT payments because they are precise and unbiased when forecasting are made. This will assist the South African authorities in decision making for future PIT revenue.