The thermal comfort of pedestrians in outdoor spaces is important to realise the high quality of urban living. Understanding outdoor wind environments at the pedestrian level is crucial to predicting convective heat loss and to evaluating the thermal comfort of the human body. Previous studies have analysed the convective heat transfer between computational thermal manikin and surrounding environments with RANS modelling, while the application of large eddy simulation (LES) has not been explored yet. Choosing a proper turbulence model is essential to accurately predict wind turbulence and convective heat transfer. In this work, we conduct LES to measure the human convective heat loss to the environment. The LES model is validated against published wind tunnel experiments and the results are compared with the ones of the RANS model. The effects of the subgrid-scale (SGS) model, inlet fluctuating algorithm, length of sampling period, time step and mesh resolution are discussed. The results of the LES model are generally in better agreement with the experimental results and reduce the relative error of average convective heat flux from 22% to 8% compared with the RANS model. The accuracies in predicting the convective heat flux at the back, pelvis, forearm/hand and thigh are significantly improved. In addition, selecting an appropriate subgrid scale model is crucial to predict the convective heat flux. DSL and WALE model can obtain more accurate results, while SSL model overpredicts the heat flux and has worse performance than RANS. The study demonstrates the feasibility of the LES model in predicting the convective heat transfer of the human body in outdoor environments, which provides a guideline and would assist in better assessing outdoor thermal comfort in future studies.