Haloketones (HKs) is one class of disinfection by-products (DBPs) which is genetically toxic and mutagenic. Monitoring HKs in drinking water is important for drinking water safety, yet it is a time-consuming and laborious job. Developing predictive models of HKs to estimate their occurrence in drinking water is a good alternative, but to date no study was available for HKs modeling. This study was to explore the feasibility of linear, log linear regression models, back propagation (BP) as well as radial basis function (RBF) artificial neural networks (ANNs) for predicting HKs occurrence (including dichloropropanone, trichloropropanone and total HKs) in real water supply systems. Results showed that the overall prediction ability of RBF and BP ANNs was better than linear/log linear models. Though the BP ANN showed excellent prediction performance in internal validation (N25 = 98–100%, R2 = 0.99–1.00), it could not well predict HKs occurrence in external validation (N25 = 62–69%, R2 = 0.202–0.848). Prediction ability of RBF ANN in external validation (N25 = 85%, R2 = 0.692–0.909) was quite good, which was comparable to that in internal validation (N25 = 74–88%, R2 = 0.799–0.870). These results demonstrated RBF ANN could well recognized the complex nonlinear relationship between HKs occurrence and the related water quality, and paved a new way for HKs prediction and monitoring in practice. Display omitted •Linear/log linear regression models cannot well predict haloketones (HKs) levels.•Back propagation (BP) is good to predict HKs in internal but bad in external validation.•Radial basis function artificial neural network (RBF ANN) well predicts HKs levels.•RBF ANN can well recognize complex relationships between HKs and water quality.