The goal of this study was to use naturalistic driving data to characterize the longitudinal and lateral accelerations of vehicles making a left turn from a stop at signalized intersections. Left turn maneuvers at 15 intersections were extracted from the Second Strategic Highway Research Program (SHRP2) database. A subset of 420 traversals for lead vehicles that were initially stopped and negotiated their left turns unimpeded by oncoming traffic was used for the analysis. For each traversal, we extracted information regarding the driver’s sex and age, the vehicle type, the vehicle’s longitudinal and lateral acceleration, and on-board forward-facing video. From the video, we further extracted information about whether the road was dry/wet and if it was day/night, and from aerial photographs of the intersections we extracted the radius of each left turn path through the intersection. From pooled data across all intersections, we identified a 3-phase longitudinal-acceleration profile where the acceleration increased from zero to an initial peak of about 0.255 g by 1.706 s after onset, decreased to a local minimum of about 0.055 g by 4.440 s, and then gradually increased to 0.065 g by 10 s. Three phases were also identified in the lateral acceleration, wherein onset of the lateral acceleration was delayed compared to the longitudinal acceleration, peaked at 0.315 g about 6.565 s after longitudinal acceleration onset, and then gradually decreased to 0.052 g by 10 s. Linear regression models showed that turn radius, driver age and nighttime had the most widespread effects on the amplitude and timeline of the longitudinal and lateral accelerations. Wet roads, vehicle type and driver sex had only a limited effect. This research provides valuable insight into the typical acceleration profiles of left-turning vehicles and the linear regression models developed here can be incorporated into accident reconstruction analyses of left-turning vehicles.