The purpose of this research was to identify combinations of behavioural (e.g., long gambling sessions or quitting only at closing time), physiological (e.g., getting the shakes or feeling nauseous) and emotional (e.g., depression or anger) responses to gambling, which could be used to identify problem gamblers with a high degree of confidence. In a survey of 711 regular Video Lottery Terminal (VLT) gamblers in Nova Scotia, respondents self reported the frequency with which they exhibited a list of indicators while gambling. The occurrence of these cues was then weighted by frequency per trip and number of trips to VLT locations per month in order to create a dataset reflective of the frequency of these events in the venues and their association with problem gamblers. Association analysis was then used to derive combinations of two or three cues that would identify problem gamblers with a high degree of confidence. A large number of highly predictive cue combinations were identified. Using cue combinations of up to three cues, with at least one visible cue and confidence values greater than 90%, 86.0% of the problem gamblers could possibly be identified. The average occurrence of "false approaches" would be 6.0%. The research is the first of its kind to provide a methodology and results that indicate that gamblers could be identifiable on site.