Temperature observations, especially those observations from highly urbanized areas, are most likely to fail persistence test and step test when basic quality assurance (QA) procedures are applied. According to WMO, doubtful data from the persistence test and the step test are called dead band suspected data and inconsistent suspected data. Studying these suspected observations may contribute to validating important doubtful data and improving the performance of temperature QA system. Basic QA procedures are applied to hourly temperature observations at 18 automatic weather stations in a highly urbanized area (Shanghai Expo) collected from May 2010 to April 2011. The distribution of suspected data and possible causes are investigated. The results show that temperature dead band suspected cases occur mostly at isolated stations in the winter evening. They are more likely observed at shielded and blocked stations in cloudy weather. The sensible heat flux is low in these situations, which may be the reason that temperature keeps the same value for several hours, even lasting for 11 h at most. In addition, the temperature inconsistent suspected cases can be grouped into two subsets: temperature jump cases and temperature slump cases. The temperature jump cases mostly occur around sunrise in autumn and winter while the slump cases mainly are found in the afternoon or evening of springsummer season. Moreover, the temperature jump cases appear at isolated stations, but temperature slump cases cooccur at different observing sites. The occurrence of these cases is closely related to weather conditions. As solar altitude increases at sunrise, the sheltered stations are suddenly exposed to the sun and warm up dramatically. Similarly, temperature goes down rapidly around sunset at the sheltered stations and temperature slump cases occur. In addition, the shorttime heavy precipitation also causes dramatically cooling at stations. Therefore, these suspected data are reasonable. They might be important observations in studying extreme weather events as well as environment effects on temperature. Multiple crosschecks are required when some samples fail in the basic QA test.