Because of its importance and frequency of occurrence, considerable effort has been devoted to the modeling of subcritical surface crack growth in recent years. Due to the geometrical complexities involved, available solutions are generally constrained to elastic numerical models implying self-similar crack growth. By selecting specific geometries (known as ‘benchmark’ geometries) and solving for stress-intensity factor distributions, several of these numerical models have recently exhibited reasonable agreement by assuming semi-elliptic crack shapes (see Ref 6). Over the past decade, the first author and his colleagues have evolved an experimental technique which couples the field equations of linear elastic fracture mechanics with the frozen stress photoelastic method for generating natural crack shapes and their corresponding stress-intensity distributions where neither are known a priori. The present paper focuses upon the application of this technique to two surface crack geometries approximating benchmark geometries and comparing with results from the recent literature in order to assess the validity of the semi-elliptic crack shape assumption for the surface crack in numerical models and to quantify any observed deviations. Results show that shallow crack shapes (a/2c ≈ 0.30 a/T ≈ 0.30) are accurately described by the semiellipse assumption, and the experimental stress-intensity distributions seem to be predicted with reasonable accuracy in regions of highest values. The deeper cracks, however (a/2c ≈ 0.30 a/T ≈ 0.75), exhibit a deviation in shape from a quarter-ellipse in the form of bulging near the points of intersection of the flaw border with the plate front surface, and the accompanying maximum stress intensities are some 25 percent higher than the values predicted in Ref 6 using semi-elliptic crack models with the same aspect ratio and a/T.