In situ feeding assays implemented with transplanted crustacean gammarids have been claimed as promising tools for the diagnostic assessment of water quality. Nevertheless the implementation of such methodologies in biomonitoring programs is still limited. This is explained by the necessity to improve the reliability of these bioassays. The present study illustrates how modelling the influence of confounding factors could allow to improve the interpretation of in situ feeding assay with Gammarus fossarum. We proceeded in four steps: (i) we quantified the influence of body size, temperature and conductivity on feeding rate in laboratory conditions; (ii) based on these laboratory findings, we computed a feeding inhibition index, which proved to be robust to environmental conditions and allowed us to define a reference statistical distribution of feeding activity values through the data compilation of 24 in situ assays among diverse reference stations at different seasons; (iii) we tested the sensitivity of the feeding assay using this statistical framework by performing 41 in situ deployments in contaminated stations presenting a large range of contaminant profiles; and (iv) we illustrated in two site-specific studies how the proposed methodology improved the diagnosis of water quality by preventing false-positive and false-negative cases mainly induced by temperature confounding influence. Interestingly, the implementation of the developed protocol could permit to assess water quality without following an upstream/downstream procedure and to compare assays performed at different seasons as part of large-scale biomonitoring programs.