Outbreaks of the gypsy moth,Lymantria dispar(L.), typically occur over large areas but are difficult to predict. Most gypsy moth management programs base suppression decisions on models that predict defoliation from preseason counts of egg masses in a given stand. In this study we developed a statistical model that used spatially stratified egg mass samples to predict gypsy moth defoliation on a regional scale, rather than on a stand level. The model was developed from historical defoliation sketch-map data and counts of gypsy moth egg masses under burlap bands at irregularly distributed plots in Massachusetts. These counts were used to generate interpolated surfaces of egg mass counts in grid cells (2 by 2 km) throughout the state. Maximum-likelihood procedures were used to parameterize a logistic regression model that predicted the probability of defoliation in each grid cell as a function of interpolated egg mass counts, the presence of defoliation in the previous year, and the 30-yr frequency of defoliation. Predicted probability surfaces tended to align mostly with the distribution of actual defoliation in each year. The model appeared to perform better than a previous model that was based on three-dimensional kriging of defoliation.