Local and member instability modes in structural steel members are influenced by initial geometric imperfections and residual stresses among other parameters. The robustness of stability-related design provisions and finite element modeling procedures of steel members relies on the accuracy of the accounted residual stresses. Numerous residual stress models exist in literature. However, they are based on experimental work that dates to early 1950s. In this paper, the validity of available residual stress models is assessed based on a dataset of physical residual stress measurements of more than 80 wide flange cross sections. This dataset was complemented with measurements on five additional wide flange cross sections. Collectively, it is found that the accuracy of available residual stress models, such as the commonly used ECCS model, is not sufficient in most cases. For this reason, we propose a residual stress model that was derived based on a constrained optimization problem between the experimental observations and an assumed quadratic residual stress distribution. The mean error between the proposed model and all observations reduces by 70% compared to available models in the literature. Moreover, the corresponding variance reduces by up to four times.