Despite a long and productive history as a focus of research interest, the details of how humans detect motion in an image remain controversial. This debate has not been helped by the lack of a clear parametric description of motion discrimination for some of the more simple visual stimuli employed in the literature to date. With this in mind, in the present work, we examined a peculiarity observed in the perception of the motion of second-order (contrast-modulated) stimuli: Under certain stimulus conditions, there is a reversal in the perceived direction of motion of the pattern. The aim was to quantify this phenomenon, relate the reversal to forward (veridical) and ambiguous motion, and place the behavioral data in the context of the window of visibility model of spatiotemporal contrast sensitivity. The direction of motion of contrast-modulated patterns was measured as a function of temporal frequency and carrier contrast, under different critical stimulus conditions. The stimulus properties manipulated were spatial frequency, spatial-phase relationship of carrier and sidebands, color, duration, and, most critically, the retinal location of the stimulus. On a purely empirical basis, the data reconciled several conflicts in the recent literature. From a theoretical standpoint, the data were well explained by the window of visibility approach in the majority of conditions and were partially explained in the remaining conditions. The results raise some interesting questions about underlying motion detection mechanisms and the assumptions embodied in our approach to motion modeling and the visual system in general. Supplemental materials for this article may be downloaded from app.psychonomic-journals.org/content/supplemental.