This paper introduces a weighted context-based step length estimation algorithm for pedestrian dead reckoning. Six pedestrian contexts are considered: stationary, walking, walking sideways, climbing and descending stairs, and running. Instead of computing the step length based on a single context, the step lengths computed for different contexts are weighted by the context probabilities. This provides more robust performance when the context is uncertain. The proposed step length estimation algorithm is part of a pedestrian dead reckoning system which includes the procedures of step detection and context classification. The step detection algorithm detects the step time boundaries using continuous wavelet transform analysis, while the context classification algorithm determines the pedestrian context probabilities using a relevance vector machine. In order to assess the performance of the pedestrian dead reckoning system, a data set of pedestrian activities and actions has been collected. Fifteen subjects have been equipped with a waist-belt smartphone and traveled along a predefined path. Acceleration, angular rate and magnetic field data were recorded. The results show that the traveled distance is more accurate using step lengths weighted by the context probabilities compared to using step lengths based on the highest probability context.