We applied a habitat-modeling approach using logistic regression to predict the distribution and abundance of canine heartworm (Dirofilaria immitis) in coyotes (Canis latrans) throughout California. Heartworm is an arthropod-borne parasite of considerable economic and ecological importance. In California, coyotes serve as the primary sylvatic maintenance host and represent a useful sentinel for this parasite. To develop the model, we used a large collection of coyote blood specimens and carcasses collected from spatially broad, yet nonrandom, locations in California. Survey data were useful in refining previous coarser models that predicted uniformly high prevalence of heartworm throughout the coastal and Sierra-Nevada foothills, by indicating variability within this broadly defined plant-climate zone. Due to the non-random nature and large spatial scale of our data-set, we restricted variables to those thought to be most generally important. Modeling indicated that woodlands with a relatively dense canopy, suitable breeding and host-seeking habitat of the western treehole mosquito (Ochlerotatus sierrensis), were a good predictor of heartworm prevalence. Within this habitat, prevalence increased with precipitation, which likely affected mosquito abundance. The distribution of heartworm was limited to areas with average cumulative temperatures high enough to enable larval development of heartworms within their mosquito vectors. The prediction accuracy of our model was supported by goodness-of-fit tests, cross-validation tests and external validation tests. The model provided a useful guide to the relative risk of heartworm exposure in California, although the resolution was necessarily coarse and prevalence estimates related to risk in an ordinal manner only.