Shallow landslides on steep slopes often mobilize as debris flows. The size of the landslide controls the initial size of the debris flows, defines the sediment discharge to the channel network, affects rates and scales of landform development, and influences the relative hazard potential. Currently the common practice in digital terrain-based models is to set the landslide size equal to an arbitrarily chosen grid cell dimension. Here we apply a multidimensional landslide model that assumes that a soil block fails as a rigid mass when downslope forces overcome cohesive and frictional resistances developed over the area of the slide base and lateral walls. We find that for a specified block width and length, there is a range of soil depth at which failure may occur. For low lateral root strength, basal root cohesion prevents shallow failures and lateral earth pressure prevents deep failures. As lateral root strength increases, the range of instability narrows until all soil depths are stable. We propose that in the field, failure location and size are largely controlled by the spatial structure of soil depth, topography, vegetation and shallow subsurface flow (and corresponding pore pressure development). To explore how this structure affects slope stability, we use a stochastic soil production model coupled with a nonlinear slope dependent flux model to predict the spatial variation in soil depth across an inclined planar hillslope, and then model how this depth variation affects the location and size of soil landslides. We use a search algorithm to identify the shapes and sizes of all soil blocks that would fail, as well as the least stable block. Our model is a first step toward a procedure that uses the spatial structure of controlling factors to predict landslide size, as compared to the current practice of assigning sizes based on grid resolution.