The scaling of peak flows associated with a probability of exceedance (Qp) or a specific rainfall‐runoff event (QR) with respect to drainage area (A) is known as flood scaling and it has been widely used in peak flow regionalization. The attenuation and aggregation processes within the hillslopes and river network in a rainfall‐runoff event, provide a framework to test the scaling of QR. Although scaling of Qp has been reported in empirical studies, its physical interpretation is compromised, since Qp at each site could come from different rainfall‐runoff events. To address this problem, the authors explored the effect of actual variabilities of rainfall and soil moisture fields, and the effect of the river network structure, in the scaling of peak flows of 85 rainfall‐runoff events and peak flow quantiles that were observed in the Iowa River Basin at 43 streamflow gauges. The authors established empirical evidence that addresses two questions: (1) What does control the performance of the scaling of observed QR? (2) What is the interplay between sampling errors and the selection of explanatory variables in the construction of regional regression models for QR and Qp? For the first question, the authors found that the slope magnitude in the scaling of the rainfall intensity fields with respect to A controls the scaling' performance of QR. Regarding the second question, the authors demonstrate that the inclusion of river network descriptors should improve the regional equations to estimate peak flow quantiles unless stream gauging sampling errors affect the analysis. Key Points The performance of the scaling of peak flows from rainfall‐runoff events is explained by the scaling structure of rainfall intensity fields The river network structure represented by width function descriptors can improve the regional equations to estimate peak flow quantiles The selection of width function descriptors in regional equations is hampered by sampling errors and limited number of streamflow gauges