Dot comparison tasks are a prominent method for investigating non-symbolic numerical processing and its development during childhood. In these tasks participants must select the more numerous of two dot arrays presented to them. Until now, it has been assumed that the developmental improvement observed in the dot comparison task during childhood is due to increasingly more precise internal numerical representations. However, this sharpening hypothesis was recently challenged by Piazza et al. (2008), who proposed that the improvement is governed by a so-called filtering process, i.e. children become better at filtering out irrelevant stimulus dimensions, such as the size of the dots, or their convex hull (smallest contour around all dots). But the filtering hypothesis is difficult to reconcile with the finding, that that different visual cues can influence numerical comparisons in opposite ways, as shown by our previous research. Whereas for convex hull, dot arrays with larger convex hulls are judged as more numerous, for dot size, dot arrays with smaller dots are judged as more numerous. In the current study we aimed to investigate whether filtering and sharpening processes affect convex hull and dot size differently during typical development. To this end, we implemented a dot comparison task in which dot size and convex hull are manipulated separately and compared performance and congruency effects between adults and 10-year-old children. Our results suggest that depending on the stimulus dimension at hand (convex hull vs. dots size), the increase in performance can be explained by both sharpening and filtering.