Based on GPS readings taken four mornings running, this article uses statistical methods, taking account of observation dependence, to study bias and variance of errors. It is shown that with or without filtering based on criteria such as the number of satellites or PDOP, mean longitude and latitude values are unbiased. In the experiment carried out, filtering was used to improve the precision of longitude readings, but had a lesser effect on latitude. However, filtering can cause long delays before obtaining a reading. Replicating readings can increase their precision, provided certain operating conditions are respected.