Many business cycle indicators present asymmetric features that have long been recognized in economics. Basically the contraction periods in an economy are more violent but also more short-lived than the expansion periods, where the dynamics of GDP/GNP growth present asymmetric cyclical developments with upswings which last longer than downturns. Nonlinear models are therefore required to capture the features of the data generating mechanisms of such macroeconomic business cycle series, since linear models are incapable of generating such behaviour. Relying on the smooth transition autoregressive (STAR) model, one of the many non-linear models developed in the nineties, this paper presents empirical evidence in favour of the proposition that the dynamic behaviour of GDP changes over the business cycle. Using quarterly growth rates for seasonally unadjusted GDP data, both, for the Euro-area and Portugal we uncover evidence in favour of asymmetric behaviour for these variables. The nonlinear features are empirically far more evident for the case of Portugal and less sticking so for the GDP for the Euro-area. These nonlinear features show up not only in the form of distinct impulse response functions calculated from different starting points for the shocks, but are also apparent in the completely different dynamics displayed by the two extreme regimes that characterize the "recession" periods and the "high" growing phases of the business cycle.