The simplest expression of the urban surface is alternation of buildings and streets. The space above a street between buildings is called an urban canyon. Air circulation in an urban canyon, which affects the behavior of air pollutants emitted by motor vehicles, is a significant factor in investigating the urban environment. However, because of the difficulty in installing observation equipment, few observational studies on air circulation have been conducted in actual urban canyons. In the present study, in order to detect three-dimensional air circulation in an urban canyon quantitatively, three wind vector components were measured using interval camera and tracer balloons with a known ascending rate. The three-dimensional position of the tracer balloons was determined by microscopic measurements of the position and diameter of the balloons on the negatives taken at an interval of two seconds (Fig. 2, Photo 1). Thereafter, wind vectors that drive the tracer balloons were calculated by subtracting the ascending rate of the respective balloons from the vectors of balloon movement. The direction and length of the target canyon, located on the former campus of Hiroshima University in Hiroshima City, are NNW-SSE and about 80 m, with canyon height and width of about 18 m and 25 m, respectively (Fig. 1). During the observation periods, wind at 3 m above the rooftop level blew obliquely to the canyon direction, and wind velocity ranged from 1.6 to 2.3 m/s (Table 2). The results of observations can be summarized as follows: 1) Measurement of air circulation using tracer balloons and interval camera is a simple method for the set-up of observation equipment, and can be used without serious error (Table 1). Therefore, this method is applicable to actual urban canyons. 2) Although there were marked contrasts in the distribution of air and surface temperatures in a canyon, no significant difference was observed in air circulation between approximately noon and during the afternoon (Fig. 4). Therefore, the air circulation detected in this study could be induced mechanically by downward air motion due to the blocking effect of leeward buildings. Air flow parallel to the canyon direction was in a spiral stream (Fig. 3). Moreover, the intensity of circulation defined as wind vorticity (Fig. 5) in the cross-section of the canyon correlated positively and significantly with the wind component perpendicular to the canyon direction at 3 m above the rooftop level (Fig. 6). 3) Wind velocity parallel to the canyon direction increased toward the leeward canyon direction (Fig. 3 (b), (c)). The increase in the wind component rate parallel to the canyon direction was nealy proportional to the downward wind velocity at the canyon top (Fig. 7), i. e., the net downward air flow from the canyon top should accelerate wind velocity in the canyon along the canyon direction. In conclusion, it was observationally confirmed that air circulation in the urban canyon spirals with descending and ascending motions on the leeward and windward sides of the canyon versus to the wind aloft, respectively. Moreover, the intensity of air circulation varies according to the wind velocity aloft, and wind velocity parallel to the canyon direction is accelerated by net inflow from the canyon top. Since the wind conditions in the present observation were rather limited, further observations are needed for various wind velocities and directions.