Research demonstrating that infants discriminate between small (e.g., 1 vs. 3 dots) and large numerosities (e.g., 8 vs. 16 dots) is central to theories concerning the origins of human numerical abilities. To date, there has been no quantitative meta-analysis of the infant numerical competency data. Here, we quantitatively synthesize the evidential value of the available literature on infant numerosity discrimination using a meta-analytic tool called p-curve. In p-curve the distribution of available p-values is analyzed to determine whether the published literature examining particular hypotheses contains evidential value. p-curves demonstrated evidential value for the hypotheses that infants can discriminate between both small and large unimodal and cross-modal numerosities. However, the analyses also revealed that the published data on infants' ability to discriminate between large numerosities is less robust and statistically powered than the data on their ability to discriminate small numerosities. We argue there is a need for adequately powered replication studies to enable stronger inferences in order to use infant data to ground theories concerning the ontogenesis of numerical cognition.