Daphnia magna is a bioindicator organism accepted by several international water quality regulatory agencies. Current approaches for assessment of water quality rely on acute and chronic toxicity that provide no insight into the cause of toxicity. Recently, molecular approaches, such as genome wide gene expression responses, are enabling an alternative mechanism based approach to toxicity assessment. While these genomic methods are providing important mechanistic insight into toxicity, statistically robust prediction systems that allow the identification of chemical contaminants from the molecular response to exposure are needed. Here we apply advanced machine learning approaches to develop predictive models of contaminant exposure using a D. magna gene expression data set for 36 chemical exposures. We demonstrate here that we can discriminate between chemicals belonging to different chemical classes including endocrine disruptors and inorganic and organic chemicals based on gene expression. We also show that predictive models based on indices of whole pathway transcriptional activity can achieve comparable results while facilitating biological interpretability.