The multi-logistic regression method was used to establish a regression model for predicting oviposition site selection in the Locusta migratoria manilensis in Nandagang farm, a coastal area Hebei province. Eight potential factors are observed including vegetation species and densities, soil properties (water content at 5 cm depth, salinity, pH and organic matter) and microrelief properties (south slope and north slope) in coastal areas. Data of locust eggpods, vegetation, soil and microrelief were collected from two spatial scales in 2002 and 2003: 450m intervals throughout the study area, and 50m grids for the possible egg-laying areas where locust plague occured or vegetation was sparse. The obtained regression model includs only vegetation density (veg_d) and soil water content at 5 cm (water), logP(Y=1)1-P(Y=1)=21.63-76.23water-543log(water)-0.86(veg_d). Where P(Y=1) is the probablity of a site selected by Locusta migratoria manilensis. After evaluated by methods of Goodness of Fit, Predictive Accuracy and Model Chi-square Statistic, the established regression model is proved to be reliable and can predict eggpods occurrence well. This model can be used for locust plague prediction and early decision-making of its control.