• Fuel consumption (FC) from LDPVs is measured using on-board diagnostic method (OBD). • The FC of the OBD is 7.1% lower than that of the carbon balance results. • The discrepancy between the approved FC and real-world FC is 13%±18%. • There is a strong relationship (R2=0.984) between the average speed and relative FC. An increasing discrepancy between real-world and type-approval fuel consumption for light-duty passenger vehicles (LDPVs) has been reported by several studies. Normally, real-world fuel consumption is measured primarily by a portable emission measurement system. The on-board diagnostic (OBD) approach, which is flexible and offers high-resolution data collection, is a promising fuel consumption monitoring method. Three LDPVs were tested with a laboratory dynamometer based on a type-approval cycle, the New European Driving Cycle (NEDC). Fuel consumption was measured by the OBD and constant-volume sampling system (CVS, a regulatory method) to verify the accuracy of the OBD values. The results of the OBD method and the regulatory carbon balance method exhibited a strong linear correlation (e.g., R2 = 0.906-0.977). Compared with the carbon balance results, the fuel consumption results using the OBD were 7.1%±4.3% lower on average. Furthermore, the real-world fuel consumption of six LDPVs was tested in Beijing using the OBD. The results showed that the normalized NEDC real-world fuel consumption of the tested vehicles was 13%±17% higher than the type-approval-based fuel consumption. Because the OBD values are lower than the actual fuel consumption, using a carbon balance method may result in a larger discrepancy between real-word and type-approval fuel consumption. By means of the operating mode binning and micro trip methods, a strong relationship (R2 = 0.984) was established between the average speed and relative fuel consumption. For congested roads (average vehicle speed less than 25 km/h), the fuel consumption of LDPVs is highly sensitive to changes in average speed.