How do we remember urban space? How can we measure what is remembered? This thesis presents a new approach to study urban spatial perception in an efficient, automated, and scalable way. It explores the use of novel tools developed for online surveys and data visualization. Previous studies in human spatial perception have used methods such as face-to-face interviews, questionnaires, recognition tasks and surveys that ask subjects to draw sketch maps. Those conventional methods produced significant urban studies such as the one by Kevin Lynch (1960), but they are laborious, sensitive to the individuality of subjects, prone to examiners' biases and conducted with a limited number of subjects. Their results are also difficult to quantify. In contrast, the method developed here uses geo-tagged street views and a web-based visual survey. An online experiment conducted in this thesis collected 394 participants in 20 days who were asked to guess the locations of street views from a familiar neighborhood. Results are presented in the form of interactive visualizations. Analysis revealed that memory for exact location of place improves with degree of interaction and proximity to center, rather than number of encounters; memory for one location may vary dramatically between different viewpoints. The results also suggest that the irregularity of urban structure doesn't prevent the forming of strong mental images. While this new method cannot completely replace face-to-face interviews, it demonstrates the possibility of using available technology to scale visual surveys to hundreds or even thousands of people and rapidly visualize the resulting data. It thus opens up new possibilities for large-scale, fine-grained studies in urban perception.