Storm-petrels have been shown to use dimethyl sulfide (DMS) as a foraging cue, suggesting that this compound may be used to predict their distribution. We describe a new distribution model that employs machine learning software and geographic information systems to model storm-petrel distribution. We used environmental predictor variables that included newly available climatologies of sea surface DMS concentrations to construct distribution maps of fork-tailed storm-petrel (Oceanodroma furcata) and Leach’s storm-petrel (O. leucorhoa) in the North Pacific and Bering Sea. Model accuracy was assessed by (1) using the area under the receiver operating characteristic curve (AUC) values and (2) comparing predicted distributions to presence and non-detection data from two opportunistic pelagic surveys performed in summer 2008. Models using all predictor variables gave AUC values of 0.89 and 0.75, sensitivity values of 0.73 and 0.61, and specificity values of 0.83 and 0.73 for fork-tailed and Leach’s storm-petrel, respectively. Models using all predictor variables except DMS gave AUC values of 0.87 and 0.74, sensitivity values of 0.81 and 0.60, and specificity values of 0.77 for fork-tailed and Leach’s storm-petrel, respectively. The large-scale link between DMS and how storm-petrels use it to locate foraging areas was reinforced by the partial dependence of DMS on the relative index of occurrence (RIO) of storm-petrels, and by a decrease in AUC values when removing DMS as a predictor. This work is a preliminary step toward linking seabird distribution to globally important infochemicals and should be a basis for further study.