Improving the quantification of the transverse mixing coefficient in rivers is essential for a better estimation of pollutant dispersion. This coefficient is evaluated from tracing experiments or from semi-empirical equations assuming uniform flow conditions. This assumption, verified for laboratory experiments, is questionable for river studies. For example, piedmont rivers exhibit nonuniform flows due to their longitudinal morphological variations. This work aims at studying the longitudinal variability of the transverse mixing coefficient in a river with longitudinally varied morphology and the overlooked impact of the flow nonuniformity on the coefficient's values. Tracing experiments were conducted in the Durance River in France using slug injections and concentration measurements made at the cross-sections delineating homogeneous sub-reaches. Using the well-known 1D transverse diffusion model, values of the transverse mixing coefficient are determined for all consecutive sub-reaches. These values substantially vary along the studied river, with stronger mixing efficiency at pools and complex riffles and weaker mixing efficiency in straight sub-reaches with emerging obstacles. The dimensionless transverse mixing coefficient is strongly related to a parameter quantifying the deviation from flow-uniformity, i.e., indicating flow acceleration, deceleration or a succession of both within the sub-reaches. While usual formulas from the literature fail to predict our empirical results, a new equation including the flow nonuniformity parameter fits our experimental data with high accuracy. After further validation, such equation may be used for predicting the transverse mixing coefficient in rivers with longitudinally varied morphology.