Prediction of species distribution and abundance has long been a goal of community ecology.  Multiple statistical techniques (e.g. linear regression, neural networks) have attempted to make these predictions from environmental variables.  Bayesian Belief Networks have potential for use in this field because they incorporate multiple environmental gradients, multiple species, are hierarchical in structure, and integrate previous knowledge about species-environment relationships.  We used the Netica^TM (Norsys Systems Corporation, Vancouver, British Columbia) software to construct empirical relationships that describe how abundance varies with trait composition for 102 taxa.  The Bayesian network used environmental variables from 140 sites in Oregon and Washington to create the probability of occurrence for 59 trait states (e.g. multivoltine, small body size).  The specific combination of traits for each taxon was then used to estimate the taxon’s probability of categorical abundance (e.g. 0, <10, or >50 individuals at a site).  While the program performed fairly well in predicting the abundance of some taxa from environment-trait relationships, we could not adequately explain the low abundances of some naturally rare species.   Additionally, the Netica software required that environmental and abundance values be categorical, decreasing the accuracy of our models.  Alternative Bayesian modeling approaches that address these limitations will be discussed.