Although upper body posture is a very important factor in cycling in terms of aerodynamic drag and body fatigue, there is no easy way for a cyclist to know his or her upper body posture. Therefore, we propose a method to estimate the upper body posture using time series information of the load applied to the handlebar. We developed a posture estimation model using Long Short Term Memory (LSTM) with the load applied to the handlebar and the cyclist’s skeletal information as inputs and the joint angles of the upper body as outputs. The accuracy of the model was improved compared to that without time series information. To estimate the posture of the upper body while riding outdoors, a measurement system was developed for an outdoor environment. The upper body posture was estimated based on the measured loads. The results of the estimation were evaluated qualitatively.