A long-standing classification problem in archaeology is determining the type of weapon delivery system used by people in the past. This is usually done by comparing archaeological points to known dart and arrow points from the ethnographic and archaeological record. There are no simple criteria to discriminate between these two states and the challenge is to identify a subset of traits and their interactions to solve this problem. Here we introduce a Bayesian technique of classifying dart and arrow. Using machine-learning feature selection, we first find the optimal set of variables for classification. We then use a Generalized Additive Model to model the interaction of these variables in a Bayesian logistic framework to capture the nonlinear decision boundary between darts and arrows and assign probabilities of a point belonging to either state. To counteract the imbalance of having more arrows than darts, we adjust the typical decision cutoff using an iterative approach that balances sensitivity and specificity. We increase the sample of known arrow and dart points with 102 previously published specimens from the West. The code for our model is available and easily accessible through an online application. We apply our model to published dart-versus-arrow classifications to demonstrate its utility.