Visual crowding is the inability to perceive properly peripheral stimuli within clutter. Previous work has shown that crowding is affected by perceptual grouping: when the flankers do not group with the target, crowding decreases, leading to uncrowding. Typically, the grouping is driven by Gestalt-like effects. Here, we hypothesized that grouping by statistical properties can affect crowding as well. Participants performed a Vernier acuity task in the periphery. Vernier stimuli were presented in isolation (baseline), surrounded by one square (crowding baseline), or within multi-element displays made of 35 flanker squares with varying tilt. We manipulated the tilt of each square according to different types of statistical distributions: one-peak narrow (low variance around a single mean), one-peak wide (larger variance), two-peak (two highly separable narrow distributions), and uniform. The tilt of the central square directly surrounding the Vernier target was fixed at either the mean or an outlier of the distribution. In a separate adjustment task, we measured participants' precision in estimating the mean tilt with the same distribution types. Our results show an inverse relationship between ensemble summaries and crowding: the larger the precision in statistical representations, the smaller the effect of crowding. Crowding was reduced when the central flanker corresponded to the mean of a distribution represented with higher precision (e.g., one-peak narrow), whereas it increased when it was an outlier. We discuss the role of statistical representations and the potential effect of other factors, such as the overall variance and the heterogeneity of flanker stimuli, in visual crowding.