Regular aerosol backscatter measurements using an elastic-backscatter lidar were performed between May 2000 and November 2002 in Barcelona (Spain), in the frame of EARLINET (European Aerosol Research Lidar Network). The mixing layer height, required to understand the chemical and physical processes taking place in the low troposphere, was one of the major parameters to be retrieved. Three analytic definitions of the ML height have been tested using the range squared-corrected lidar signal: (1) the minimum of its first derivative, (2) the minimum of its second derivative, and (3) the minimum of the first derivative of its logarithm. The strong coastal and orographic influences and the climatological settling of Barcelona determine the complexity of its atmospheric boundary layer dynamics and the high heterogeneity of the lidar signals. Therefore, single lidar analyses do not allow an unambiguous determination of the mixing layer height in many cases and complementary data are needed, such as synoptic maps, backtrajectories, radiosoundings and solar irradiance profiles. The resulting mixing layer heights were compared to radiosoundings, and the second method was found to give statistically the best results. This definition was used to process the whole dataset. A number of 162 days and 660 profiles were examined. The mixing layer height was inferred in cases such as low clouds, Saharan dust events and sea breeze and mountain induced recirculation. Variations between 300 and 1450 m were observed over the three years.