Non-contact measurement techniques using laser scanning have the advantage of fast acquiring large numbers of points. However, compared to their contact-based counterparts, these techniques are known to be less accurate. The work presented in this paper aims at improving the accuracy of these techniques through an error correction procedure based on an experimental process that concerns mechanical parts. The influence of the three parameters defining the relative position and the orientation between the sensor and the surface is studied. The process used to build an experimental global model of error is presented and applied to a typical part composed of planes or skewed surface. The systematic errors have been reduced by half in comparison to the reference values while the random errors have slightly increased. This phenomenon is due to the fact that the errors correction model does not take into account the local response of the laser sensor. A second model taking into account the local defect has been developed. Its application to an example of inspection of a mechanical part shows an improvement of the results of the correction.