An extensive statistical analysis is proposed to determine the best relevant roughness parameters as a function of spatial scales that affect adhesion on surfaces. The methodology is based on a multiscale decomposition of the roughness surface and is linked with adhesion measurements. This method is applied to study cell adhesion on a very wide range of roughnesses of titanium substrates (22 surfaces, the average roughness R-a from 1 to 21 mu m) tooled by an electro-erosion process and coated with a polyelectrolyte, that leads to identical surface chemistry. It is shown that the scale length of observation should be a few times the cell size to put into evidence the influence of the surface morphology on cell adhesion. It is observed that the adhesion is the lowest when the distance between the asperities of the roughness is near the cell size.