Low concentrations of pollutants may already be associated with significant health effects. An accurate assessment of individual exposure to pollutants therefore requires measuring pollutant concentrations at the finest possible spatial and temporal scales. Low-cost sensors (LCS) of particulate matter (PM) meet this need so well that their use is constantly growing worldwide. However, everyone agrees that LCS must be calibrated before use. Several calibration studies have already been published, but there is not yet a standardized and well-established methodology for PM sensors. In this work, we develop a method combining an adaptation of an approach developed for gas-phase pollutants with a dust event preprocessing to calibrate PM LCS (PMS7003) commonly used in urban environments. From the selection of outliers to model tuning and error estimation, the developed protocol allows to analyze, process and calibrate LCS data using multilinear (MLR) and random forest (RFR) regressions for comparison with a reference instrument. We demonstrate that the calibration performance was very good for PM1 and PM2.5 but turns out less good for PM10 (R2 = 0.94, RMSE = 0.55 μg/m3, NRMSE = 12 % for PM1 with MLR, R2 = 0.92, RMSE = 0.70 μg/m3, NRMSE = 12 % for PM2.5 with RFR and R2 = 0.54, RMSE = 2.98 μg/m3, NRMSE = 27 % for PM10 with RFR). Dust events removal significantly improved LCS accuracy for PM2.5 (11 % increase of R2 and 49 % decrease of RMSE) but no significant changes for PM1. Best calibration models included internal relative humidity and temperature for PM2.5 and only internal relative humidity for PM1. It turns out that PM10 cannot be properly measured and calibrated because of technical limitations of the PMS7003 sensor. This work therefore provides guidelines for PM LCS calibration. This represents a first step toward standardizing calibration protocols and facilitating collaborative research.