In the 1950s, NACA conducted a series of low-speed cascade experiments investigating the performance of NACA 65-series compressor cascades with tests covering multiple airfoils of varying camber and with variations in solidity and air inlet angle. Most of the configurations show transition via laminar separation—both on suction and pressure side—characterized by a relatively flat region in pressure distribution, while turbulent reattachment is characterized by a rapid pressure recovery just downstream of the separated region. In the current study, wall-resolved large-eddy simulation (LES) has been used to predict transition via laminar separation in such compressor configurations as well as the resulting airfoil losses. Six different cascades with local diffusion factor varying from 0.14 to 0.56 (NACA 65-010, 65-410, 65-(12)10, 65-(15)10, 65-(18)10, and 65-(21)10 cascades) were analyzed at design conditions. In addition, the loss bucket for various angles of attack off-design conditions has been computed for the NACA 65-(18)10 cascade. Chord-based Reynolds number for all the experiments considered here was held at 250,000. This allows sufficient grid resolution in these LES analyses at an acceptable computational cost, i.e., up to 20,000 CPU hours per case. Detailed comparisons to test data are presented in the form of surface pressure coefficient, drag coefficient, losses, and momentum thickness ratio. The results show that LES is capable of capturing transition via laminar separation relatively well for most of the cases, and consequently, may constitute a predictive tool for assessing losses of different compressor airfoils.