This paper considers using nonnormal error distributions for the hedonic wine price function, to predict 'average' wine prices and hence identify over and under-priced wines. In addition to the log-normal distribution, the Weibull, gamma, exponential and log-logistic distributions are employed for a comprehensive data set of over 6,000 Australian wines. Estimates of marginal attribute impacts do differ with distribution employed with the Weibull being the most different. On average for all wines, the log-logistic distribution produces the most accurate predictions, however, the performance of distributions differs according to price range. Combining the predictions can produce substantially lower absolute errors on average, because of the varying accuracy performance of the distributions.