Selection of galaxy clusters by mass is now possible due to weak gravitational lensing effects. It is an important question then whether this type of selection reduces the projection effects prevalent in optically selected cluster samples. We address this question using simulated data, from which we construct synthetic cluster catalogues both with Abell's criterion and an aperture-mass estimator sensitive to gravitational tidal effects. The signal-to-noise ratio of the latter allows to some degree to control the properties of the cluster sample. For the first time, we apply the cluster-detection algorithm proposed by Schneider to large-scale structure simulations. We find that selection of clusters through weak gravitational lensing is more reliable in terms of completeness and spurious detections. Choosing the signal-to-noise threshold appropriately, the completeness can be increased up to 100%, and the fraction of spurious detections can significantly be reduced compared to Abell-selected cluster samples. We also investigate the accuracy of mass estimates in cluster samples selected by both luminosity and weak-lensing effects. We find that mass estimates from gravitational lensing, for which we employ the zeta-statistics by Kaiser et al., are significantly more accurate than those obtained from galaxy kinematics via the virial theorem.