The distribution of Permian alluvial fan lithofacies in a quarry at Zygmuntowka near Checiny, Holy Cross Mts., in southern Poland was investigated using ground penetrating radar (GPR) in order to create a training image for multiple point statistics (MPS) reconstructions of alluvial fan sedimentary facies. Five pseudo-3D GPR datasets were collected, processed and uploaded for interpretation into SKUA-GOCAD 3D geological modelling software. Three radar facies were distinguished based on the 3D geometrical pattern of radar reflections and linked to lithofacies described from the quarry by Zbroja et al. (1998). A statistical summary showed that ~50% of the lithofacies resulted from gravity flows (mostly non-cohesive), while the remaining proportion was deposited by unconfined and confined flash floods. Fluvial sedimentary facies left by waning of catastrophic floods or reworking during fair weather, although not prevalent, could not be distinguished from confined flood deposits based only on GPR data. The GPR datasets together with information from field observations were used to carry out MPS simulations and estimate the most probable 3D model of lithofacies at the quarry scale. This model will in turn serve as a training image for MPS reconstructions of alluvial-fan facies of Rotliegend conglomerates in the multi-scale geological model of the Gorzow Block (western Poland).