The conventional convolutional model is widely applied to generate synthetic seismic data for numerous applications including amplitude-versus-offset forward modeling, seismic-well tie, and inversion. This approach assumes frequency-independent reflection coefficients and time-invariant seismic wavelets in laterally homogeneous elastic media. We have extended the conventional convolutional model to heterogeneous poroelastic media in which reflection coe?icients are frequency-dependent and the seismic wave gets attenuated as it propagates. First, we decompose the seismic wavelet into mono-frequency components through Fourier transform. Then, to account for the attenuation effects at the reflection interfaces, we multiply the frequency-dependent reflection coe?icients series with an attenuation function of frequency-variant quality factor Q. Finally, we convolve the above product results with a mono-frequency wavelet and sum all the frequencies together to obtain the synthetic seismograms. The advantage of the proposed frequency-decomposed nonstationary convolutional model is that it takes into account the effects of attenuation on both wave reflections and propagation in attenuative media. In addition, it employs the frequency-dependent Q instead of the constant Q utilized by the traditional nonstationary convolutional model. The technique has been applied to amplitude-versus-angle-and-frequency forward waveform modeling in attenuative media, and it shows good agreement between synthetic and real data on seismic-well ties.