ABSTRACT We investigate the accuracy of weak lensing simulations by comparing the results of five independently developed lensing simulation codes run on the same input N-body simulation. Our comparison focuses on the lensing convergence maps produced by the codes, and in particular on the corresponding PDFs, power spectra, and peak counts. We find that the convergence power spectra of the lensing codes agree to $\lesssim 2{{\ \rm per\ cent}}$ out to scales ℓ ≈ 4000. For lensing peak counts, the agreement is better than $5{{\ \rm per\ cent}}$ for peaks with signal-to-noise ≲ 6. We also discuss the systematic errors due to the Born approximation, line-of-sight discretization, particle noise, and smoothing. The lensing codes tested deal in markedly different ways with these effects, but they none-the-less display a satisfactory level of agreement. Our results thus suggest that systematic errors due to the operation of existing lensing codes should be small. Moreover their impact on the convergence power spectra for a lensing simulation can be predicted given its numerical details, which may then serve as a validation test.