In this study the threshold of thermally significant transient cavitation in vivo in dog's thigh muscle was investigated as a function of frequency from 0.246 MHz to 1.68 MHz. Cavitation, evidenced by strong emission of wide band noise monitored by a hydrophone, appeared to increase the energy absorption in tissue at the focal zone of a focused ultrasound beam as measured with an embedded thermocouple. This was indicated by a significant increase in the temperature, a loss of smooth temperature rise during the 1 s sound pulse and a significant reduction in the acoustic power transmitted through the thigh. This thermal phenomenon was associated with a strong emission of wide band noise which was monitored by a hydrophone. In addition, strong echoes appeared in ultrasound images during the pulses that caused the noise emission and the thermal effect. These echoes appeared preferentially at locations where there was acoustic heterogeneity. The measured cavitation pressure amplitude threshold was found to depend almost linearly on frequency with a slope of about 5.3 MPa MHz-1. (The extrapolated static pressure threshold was 0.6 MPa.) When these measured levels are compared to those typical of clinical application, it appears that the transient cavitation can be avoided when perfusion independent high temperature hyperthermia is induced with focused and pulsed ultrasound fields. However, intensities required during scanned focused ultrasound hyperthermia, where sharply focused transducers are used to heat large tumors at low frequencies (1 MHz or below), could rise above the threshold. Thus, care should be taken when focused ultrasound systems are designed so that the maximum peak pressure is below the threshold in order to avoid unpredictable biological effects induced by transient cavitation. Finally, it is unlikely that the present diagnostic ultrasound units which operate at higher frequencies and in pulsed mode could cause transient cavitation in vivo.