While photogrammetry has become popular in archaeology and heritage management as an effective, low-cost method for generating detailed three-dimensional models, it remains to be established that the accuracy of model-derived measurements is sufficient for analytical purposes. Based on an expedient, in-field model processing protocol, we report preliminary results concerning the accuracy of artifact provenience information derived from photogrammetry models of excavation surfaces at the Upper Palaeolithic site of Shuidonggou Locality 2 in China. Error in model-derived provenience can range easily into the centimeter scale; accuracy in some spatial axes are significantly, but weakly, affected by the size of the sampled surface. While the observed error range is larger than thresholds proposed for Palaeolithic excavations, it is arguably acceptable in settings where the analytical demand for provenience precision is lower. We identify possible sources of error and discuss how model accuracy can be improved by additional systematic testing.