An important aspect of eruption forecasting is predicting the path of propagating dikes. We show how lateral dike propagation can be forecast using the minimum potential energy principle. We compare theory to observed propagation paths of dikes originating at the Bárðarbunga volcano, Iceland, in 2014 and 1996, by developing a probability distribution for the most likely propagation path. The observed propagation paths agree well with the model prediction. We find that topography is very important for the model, and our preferred forecasting model considers its influence on the potential energy change of the crust and magma. We tested the influence of topography by running the model assuming no topography and found that the path of the 2014 dike could not be hindcasted. The results suggest that lateral dike propagation is governed not only by deviatoric stresses but also by pressure gradients and gravitational potential energy. Furthermore, the model predicts the formation of curved dikes around cone‐shaped structures without the assumption of a local deviatoric stress field. We suggest that a likely eruption site for a laterally propagating dike is in topographic lows. The method presented here is simple and computationally feasible. Our results indicate that this kind of a model can be applied to mitigate volcanic hazards in regions where the tectonic setting promotes formation of laterally propagating vertical intrusive sheets.