This research utilizes the k-space computational method to design an intracavitary probe for hyperthermia treatment of prostate cancer. A three-dimensional (3D) photographical prostate model, utilizing imaging data from the Visible Human Project®, was the basis for inhomogeneous acoustical model development. The acoustical model accounted for sound speed, density, and absorption variations. The k-space computational method was used to simulate ultrasound wave propagation of the designed phased array through the acoustical model. To insure the uniformity and spread of the pressure in the length of the array, and the steering and focusing capability in the width of the array, the equal-sized elements of the phased array were 1 × 14 mm. The anatomical measurements of the prostate were used to predict the final phased array specifications (4 × 20 planar array, 1.2 MHz, element size = 1 × 14 mm, array size = 56 × 20 mm). Good agreement between the exposimetry and the k-space results was achieved. As an example, the -3 dB distances of the focal volume were differing by 9.1% in the propagation direction for k-space prostate simulation and exposimetry results. Temperature simulations indicated that the rectal wall temperature was elevated less than 2°C during hyperthermia treatment. Steering and focusing ability of the designed probe, in both azimuth and propagation directions, were found to span the entire prostate volume with minimal grating lobes (-10 dB reduction from the main lobe) and least heat damage to the rectal wall. Evaluations of the probe included ex vivo and in vivo controlled experiments to deliver the required thermal dose to the targeted tissue. With a desired temperature plateau of 43.0°C, the MRI temperature results at the steady state were 42.9 ± 0.38°C and 43.1 ± 0.80°C for ex vivo and in vivo experiments, respectively. Unlike conventional computational methods, the k-space method provides a powerful tool to predict pressure wavefield and temperature rise in sophisticated, large scale, 3D, inhomogeneous and coarse grid models.