Air pollutants emissions in urban areas are numerous and heterogeneous. Traditional monitoring techniques are restricted to a few highquality instruments missing this spatial heterogeneity. Recently, great interest has been given to low-cost sensors to better characterize these pollutants’ spatial distribution. However, calibration of these sensors often suffers from meteorological dependence and temporal drift. Artificial Neural Networks have been used to calibrate four Alphasense sensors measuring CO, NO, NO2 and O3. This analysis has shown that Multilayer Perceptron performed better in calibrating the NO, NO2 and O3 sensors than a multiple linear regression while showing similar performance to multiple linear regression for the CO Alphasense sensor. It has been shown that there is no gain in using more than two weeks of data to calibrate the sensors. The CO and O3 sensors have large temporal drift and methods have been developed to keep the calibration accurate over two months. The NO and NO2 sensors are less affected by a temporal drift.