This paper is concerned with the accuracy of temperature measurements with radio acoustic sounding system (RASS) consisting of a pulsed Doppler radar and an acoustic source, where the latter excites short monochromatic pulses. Through the use of a numerical model and middle and upper atmosphere (MU) radar experiments, we found that the accuracy is significantly affected by the acoustic and radar pulse length ratio. When the Bragg condition is not strictly satisfied, a numerical model predicted that the mean frequency shift fm of a RASS echo spectrum is detected between the Doppler-shifted frequency corresponding to the sound speed fs and the transmitted acoustic frequency fa. When the ratio is close to or larger than unity, fm becomes almost identical with fa, while fm approaches fs as the ratio decreases. RASS experiments involving the MU radar operating at 46.5 MHz (6.45-m wavelength) and an acoustic transmitter with a frequency of about 100 Hz showed that the observed characteristics of RASS echoes for various acoustic pulse lengths agreed quite well with model predictions. Although the numerical model suggested that a small value of the ratio is preferable for accurate measurement of temperature with RASS, the minimum value of the ratio was determined to be about 0.2 by taking into account the system sensitivity of the MU radar, since the RASS echo intensity decreases as the acoustic pulse length becomes shorter. When the lengths of the acoustic and radar pulses were set equal to about 60 m (18 acoustic wave cycles) and 300 m (1 μs in pulse duration), respectively, which gives a ratio of the acoustic pulse length to the radar pulse length of approximately 0.2, we were able to obtain temperature profiles at 5 to 9 km every 3 min with an accuracy of about 0.5 ¿C. ¿ American Geophysical Union 1993