Theoretical accounts of the visual number sense (VNS), i.e., an ability to discriminate approximate numerosities, remain controversial. A proposal that the VNS represents a process of numerosity extraction, leading to an abstract number representation in the brain, has been challenged by the view that the VNS is non-numerical in its essence and amounts to a weighted integration of continuous magnitude features that typically change with numerosity. In the present study, using two-alternative forced-choice paradigm, we aimed to distinguish between these proposals by probing brain areas implicated in the VNS with transcranial random noise stimulation (tRNS). We generated predictions for the stimulation-related changes in behavioural performance which would be compatible with alternative mechanisms proposed for the VNS. First, we investigated whether the superior parietal (SP) area hosts a numerosity code or whether its function is to modulate weighting of continuous stimulus features. We predicted that stimulation may affect the VNS precision if the SP role is representational, and that it may affect decision threshold if its role is modulatory. Second, we investigated whether the intraparietal (IP) area hosts a numerosity code independently of codes for continuous stimulus features, or whether their representations overlap. If the numerosity code is independent, we predicted that IP stimulation may improve the VNS but not continuous magnitude judgements. Our results were consistent with the hypotheses of a modulatory role of the SP and of the independence of the numerosity code in the IP, whereby suggesting that VNS is an emergent abstract property based on continuous magnitude statistics.