We evaluated mark–resight estimates of mountain goat (Oreamnos americanus) abundance using paintball marks at Seven Devils and Black Mountain, Idaho, USA. We marked mountain goats with recreational paintball equipment fired from a helicopter. Blue-violet dye marks were difficult to distinguish from dark, soiled areas on goats, while orange, yellow, and red oil-based paint marks were easily recognized, highly visible, and persisted for at least 71 days. Marking rates varied with equipment function and pilot effectiveness and ranged between 6.3 and 14.6 goats marked/hour of flight time. Precision of abundance estimates was reasonable with mark samples ≥51% of N̂. We analyzed a 3-occasion survey at Black Mountain using Program MARK to determine the likelihood and magnitude of potential sources of bias. Model M_b, which allows a behavioral response to capture, was best supported by the data, while model M_t, which allows time-varying capture probabilities, had reasonable support. The pattern of time-varying capture probabilities likely was a by-product of trap-shy behavior. Model M_b was 3 times more likely than model M_t to be the best approximating model, given the data. Behavioral response to marking and time-varying capture probabilities can be accommodated with >2 survey occasions. However, the strong evidence for a behavioral response implies that Lincoln–Petersen estimates of abundance are inappropriate. The set of candidate models did not include a model that would accommodate heterogeneity of individual sighting probabilities, which remains a concern with the mark–resight approach.