In order to approach properly a wide variety of issues concerning urban large passenger transport vehicles, such as the design of the bus structure, the comfort of passengers, the non-collision injury risk and the operating characteristics of the bus, detailed knowledge of the loads that the buses support when operating is required. These loads depend on numerous factors such as the geographic and urban features of the city where they operate, the type of route and the driver. All these factors provide the nature of these loads with a wide variability, and so studies based on experimentally obtained data during representative periods of operation must be developed. The main objective of the present paper is to carry out a representative characterization of the operating loads supported by large passenger transport vehicles during normal operation. It is with this aim that a study of the longitudinal accelerations and lateral accelerations to which large passenger transport vehicles are subjected was conducted over urban routes by using the data collected by the Global Positioning System. An extensive assessment of recorded data was carried out to evaluate whether the precision and the sample rate of the Global Positioning System were sufficient to characterize these accelerations accurately. To ensure that the sample was representative, data for an operation time of more than 600 h were recorded using 10 different models of large passenger transport vehicles operating over 13 different urban routes. From all the position data recorded, the instant longitudinal accelerations were calculated using second-order central differentiation, and the lateral accelerations were obtained using first-order central differentiation and the curvature radius. All the calculated accelerations were then subjected to data processing developed on an ad-hoc basis to filter the information that did not refer to accelerating manoeuvres. After this data-processing procedure, it was verified that both the lateral accelerations and the longitudinal accelerations fit normal probability distributions with a minimum margin of error (maximum differences of 0.165 m/s2 for lateral accelerations and 0.038 m/s2 for longitudinal accelerations).