Several different reconstruction algorithms have been proposed for photoacoustic tomography, most of which presuppose that the acoustic properties of the medium are constant and homogeneous. In practice, there are often unknown spatial variations in the acoustic properties, and these algorithms give, at best, only approximate estimates of the true image. The question as to which approach is the most robust in these circumstances is therefore one of practical importance. Image reconstruction by ??time reversal??-using a numerical propagation model with a time-varying boundary condition corresponding to the measured data in reversed temporal order-has been shown to be less restrictive in its assumptions than most, and therefore a good candidate for a general and practically useful algorithm. Here, it is shown that such reconstruction algorithms can ??trap?? time reversed scattered waves, leading to artifacts within the image region. Two ways to mitigate this effect are proposed.