The aim of this study is to evaluate determinants of maxillary canine impaction taking into account both canine position related variables and the pattern of facial growth. A retrospective clinical and radiographic analysis was carried out on 109 patients aged between 9 and 10 years at the time of first evaluation. At baseline, SN-GoMe angle, the interincisal angle, the canine angle ? and the canine distance d were used to characterize canine location and vertical facial growth. At the end of a two years follow up period the eruption state of each canine of each patient was recorded and accordingly classified as erupted or impacted on a clinical and radiographic basis. Univariate and multivariate statistical analyses were performed, including correlation among the studied variables and principal components analysis; several machine learning methods were also used in order to built a predictive model. At the end of the two years follow up period after the first examination, 54 (24.77%) canines were classified as impacted. Except for Angle ? values, there were no statistically significant differences between impacted and erupted canines. The studied variables were not significantly correlated, except for the SN-GoMe Angle and the distance d in the impacted canine group and the angle ? and the distance d in erupted canines group. All variables, except for SN-GoMe Angle in erupted canines, have a partial communality with the first two principal components greater than 50%. Among the learning machine methods tested to classify data, the best performance was obtained by the random forest method, with an overall accuracy in predicting canine eruption of 88.3%. The studied determinants are easy to perform measurements on 2D routinely executed radiographic images; they seems independently related to canine impaction and have reliable accuracy in predicting maxillary canine eruption.