The bulk of the literature investigating persistence properties of non-renewable resource prices series has focused on application of unit root tests. This paper contributes to the debate, applying a methodology which allows (1) robust detection of the presence and (if so) the number of changes, (2) inference on stationarity of the series, and (3) estimation of change locations. In contrast to previous papers, the analysis is carried out from the perspective of stationarity testing, incorporating quadratic trends and the possibility of smooth changes. For a classical database, we find significant evidence of trend stationarity in most of the series, suggesting that shocks are mostly of a transitory nature. Exceptions are silver and natural gas, with stationarity being rejected for all the specifications considered in the paper. Finally, the knowledge of the stochastic characteristics of the series allows robust detection of change points which appear to be related to economic events.