Elderly drivers are usually experienced and careful drivers. However, facing to the complexity of some traffic situations, they may have difficulties to perceive relevant pieces of information, make their decision and perform the right behaviour, resulting in higher involvement in certain types of accidents. It is typically the case for accidents that occur at intersections. Several studies indicate that elderly drivers are overrepresented in accidents that require turning manoeuvres, especially when Turning Left (TL). To investigate the difficulties experienced by older drivers (65 years old and more) and their accident exposure when they have to TL at a crossroad, two databases of the Gustave Eiffel University were investigated for this study: (1) a Naturalistic Database (NDB), collected among a sample of 76 elderly drivers (mean age of 74.4 years old) having performed a 28km route with an instrumented car (a Total of 2150 km covered and 64 hours of data collected) and (2) a Detailed Study of Accidents database (DSA), containing more than 1200 real cases of accidents occurred in Salon-de-Provence area (from 1992 to 2020). As a first step, a conceptual model to study driving errors and accidents risks during TL is proposed. Then, the exploitation of the 2 databases allowed us to extract 43 driving errors observed at the wheel (i.e., NDB) and 26 real cases of accidents (i.e., DSA) occurred during TL, which were respectively analysed and classified according to this conceptual model. Finally, from the joint-analysis of all these real cases of driving errors and accidents, a more comprehensive understanding of accident risks for elderly drivers during TL is proposed and discussed, according to 3 main categories of errors: related to Perception (38.5 % for DSA and 35% for NDB), to Cognition (42.3 % for DSA and 37% for NDB) or to driving Action (19.2 % for DSA and 28% for NDB).