In this paper, pilot-symbol-assisted transmission in cognitive radio systems over time selective flat fading channels is studied. It is assumed that causal and non causal Wiener filter estimators are used at the secondary receiver with the aid of training symbols to obtain the channel side information (CSI) under an interference power constraint. Cognitive radio model is described together with detection and false alarm probabilities determined by using a Neyman-Person detector for channel sensing. Subsequently, for both filters, the variances of estimate errors are calculated from the Doppler power spectrum of the channel, and achievable rate expressions are provided considering the scenarios which are results of channel sensing. Numerical results are obtained in Gauss-Markov modeled channels, and achievable rates obtained by using causal and non causal filters are compared and it is shown that the difference is decreasing with increasing signal-to-noise ratio (SNR). Moreover, the optimal probability of detection and false alarm values are shown, and the tradeoff between these two parameters is discussed. Finally, optimal power distributions are provided.