Using a Markov regime switching model, this article presents evidence on the well-known January effect on stock returns. The specification allows a distinction to be drawn between two regimes, one with high volatility and other with low volatility. We obtain a time-varying January effect that is, in general, positive and significant in both volatility regimes. However, this effect is larger in the high volatility regime. In sharp contrast with most previous literature we find two major results: i) the January effect exists for all size portfolios. ii) the negative correlation between the magnitude of the January effect and the size of portfolios fails across volatility regimes. Moreover, our evidence supports a decline in the January effect for all