GPS capable smart phones are emerging survey tools in transportation research field, especially in modeling individuals' mobility patterns. In route choice modeling, path observations need to be generated explicitly for the estimation. It is a challenge because the recorded data is not as dense or accurate as those from dedicated GPS devices. In this paper, we develop a methodology for generating probabilistic path observations from sparse and inaccurate location data, for state-of-the-art discrete route choice models. The difference of the proposed algorithm and the map matching algorithms is that instead of giving a unique matching result, the new algorithm generates a set of potential true paths, along with probabilities for each one to have been the true path. More importantly, the algorithm uses not only the topological measurement, but also temporal information (speed and time) in the GPS data to calculate the probability for observing the data while traveling on the proposed path. We emphasis traveling as a dynamic movement on a path, and model it as such in the algorithm. A short trip and two longer trips are used to analyze the performance of the algorithm on real data. Then, 19 trips recorded from a single user's cell phone are used in a preliminary study that estimates route choice behaviors using state-of-the-art discrete route choice modeling methodologies with the proposed probabilistic path observation generation algorithm. Abstract Keyword: route choice modeling, path observation generation, smart-phone data, GPS data, map matching