Data were collected during the recreational season of 2007 to test and refine predictive models at three Lake Erie beaches. In addition to E. coli concentrations, field personnel collected or compiled data for environmental and water-quality variables expected to affect E. coli concentrations including turbidity, wave height, water temperature, lake level, rainfall, and antecedent dry days and wet days. At Huntington (Bay Village) and Edgewater (Cleveland) during 2007, the models provided correct responses 82.7 and 82.1 percent of the time; these percentages were greater than percentages obtained using the previous day?s E. coli concentrations (current method). In contrast, at Villa Angela during 2007, the model provided correct responses only 61.3 percent of the days monitored. The data from 2007 were added to existing datasets and the larger datasets were split into two (Huntington) or three (Edgewater) segments by date based on the occurrence of false negatives and positives (named ?season 1, season 2, season 3?). Models were developed for dated segments and for combined datasets. At Huntington, the summed responses for separate best models for seasons 1 and 2 provided a greater percentage of correct responses (85.6 percent) than the one combined best model (83.1 percent). Similar results were found for Edgewater. Water resource managers will determine how to apply these models to the Internet-based ?nowcast? system for issuing water-quality advisories during 2008.