Most traffic sensors currently deployed along the existing freeway systems are single-loop detectors that directly measure only volume and lane occupancy. Because loop detector actuations (event data) consume extra resources to store and transfer, transportation agencies typically aggregate loop detector measurements into 20- or 30-s intervals in field traffic controllers before sending them to the Traffic System Control Center for archiving. Therefore, only aggregated volume and lane occupancy data for each aggregated interval are available for use. During the aggregation process, however, a vehicle may be present on a loop when the current aggregation interval terminates. In that case the vehicle is counted toward the following interval, but part of its scan counts is mistakenly assigned to the current interval. The incorrect number of scan counts can cause occupancy measurement errors, referred to as segmentation error (SE) in this study. This error occurs frequently and adversely affects the quality of loop data, especially when volume is low. Because single loops rely on certain algorithms for speed estimation using the aggregated measurements, SEs can result in biased speed estimates. This paper studies the impact of SE on speed estimates and proposes an improved speed estimation algorithm that is easy to implement and effective in eliminating SEs. Compared with Athol's speed estimation algorithm, widely used in practice, the new algorithm produces improved speed estimates. A thorough analysis of the frequency of SE and the level of negative impacts on speed estimation is also conducted.