Abstract: The arid region of northwest China is broad in area and complicated in topography. Precipitation varies significantly both temporally and spatially. However，rain gauges are very scarce and unevenly distributed. Few of rain gauges are installed at an elevation higher than 3 500 m above the sea level. Scarcity of the precipitation observation，especially，in the high mountain areas，brings great difficulties in hydrological study. TRMM remote sensing precipitation data is a potential substitute of the ground observation because of its wide coverage and high time- space resolutions. While the reliability of TRMM data is influenced by local climate，geography，topography，and some other factors，efforts have been attempted to evaluate and modify the TRMM data to improve its reliability for local application. This paper is focused on evaluating the detective ability of the TRMM precipitation data in arid region of northwest China. The evaluation was done through comparing rain gauge data with their corresponding TRMM grid data on daily and monthly scales from 1998 to 2010. Meanwhile，TRMM data was also evaluated with regard to warm and cold seasons. The daily ground observation data was from China Meteorological Data Sharing Service System. TRMM 3B42V6 data with the temporal resolution of 1 day and spatial resolution of 0.25°×0.25° was downloaded from NASA （National Aeronautics and Space Administration）. [FBI]（frequency bias index），[POD]（probability of detection）and [FAR]（false alarm ratio）were used to indicate the detective ability of precipitation occurrences，and the bias and correlation coefficient were used to assess the precipitation amount of TRMM data. The results indicated that performance of the TRMM data differentiated in sub-regions，warm and cold seasons，altitudes，and time scale. Regionally，TRMM overestimated the occurrence frequency of precipitation events in Hexi Corridor of Gansu Province，west of Inner Mongolia，and south of the Tianshan Mountains，and underestimated in north of the Tianshan Mountains. The False Alarm Ratio in the south Tianshan Mountains ranked the first，followed by Gansu Corridor，and the north Tianshan Mountains. TRMM performed better in warm season than in cold season due to snow covers which affected microwave signal. Performance of TRMM was affected by elevation，TRMM data tended to overestimate daily precipitation at low altitude while underestimate it at high altitude，monthly precipitation in warm season was also overestimated at low altitude while underestimated at high altitude. TRMM data gave a better estimation of precipitation monthly than daily．Among seventy-six pixels with the rain-gauges located inside，almost all of [FBI s]，[POD s] and [FAR s] for daily precipitation were far away from the optimal value，[POD s] and [FAR s] were smaller than 0.72 and higher than 0.37，respectively；correlation coefficients were almost smaller than 0.4；biases of fourteen pixels were within the range of ±10%，while the largest bias for other pixels reached to 389%.As for monthly precipitation，up to nine pixels，fourteen pixels and twenty-four pixels in which [FBI s]，[POD s] and [FAR s] reached to optimal value，respectively；correlation coefficients of sixty-seven pixels were higher than 0.4；and there were still fourteen pixels in which biases were within the range of ±10%. Nevertheless，pixels in which all the evaluating indices were within the acceptance range were very scarce. In summary，detective ability of TRMM precipitation products in arid area of northwest China was very limited，and modification to the data is necessary before it is properly used although it showed potential as a substitute of ground precipitation observation，and calibrating TRMM precipitation would be our next step.

Key words: arid region of northwest China, precipitation, TRMM, hydrology