Ion therapy has been well established for many decades owing to sharp depth-dose gradient and high cell-killing capability of ions. However, monitoring of the incident ions is needed to reduce ion range uncertainty for effective tumor treatment. Positron-emitting ion beams, known as radioactive ion (RI) beams, can be used as the optimal clinical approach for simultaneously treating and monitoring the incident ions using positron emission tomography (PET) imaging. The low intensity of RI beams, which previously was the main difficulty of their clinical use, was recently addressed by the development of accelerator technology. Since the quantity of interest in ion therapy is the dose and a direct comparison of planned and delivered doses is highly desirable, we focused on predicting the depth dose from PET images for RI beams in this work for the first time. We developed a method for predicting the depth dose of polyenergetic ion beams of 11C and 15O from PET images of these ion beams. The depth dose in water was measured for mono- and polyenergetic ion beams. The monoenergetic ion beams had a momentum acceptance (MA) of ±0.25%, while the polyenergetic beams had an MA of ±2.5%. PET imaging was performed during and shortly after irradiation of a polymethylmethacrylate (PMMA) phantom with mono- and polyenergetic radioactive ion beams. The energy distribution of each polyenergetic ion beam was obtained from the measured momentum distribution of that beam. A set of depth dose and PET profiles in PMMA was constructed for several monoenergetic ion beams using measured data of a monoenergetic ion beam and the range-energy relation. To estimate the depth dose for each polyenergetic beam, a linear combination of weighted PET profiles of monoenergetic beams was iteratively fitted to the measured PET profile of the polyenergetic beam, and the resulting weights were combined with the corresponding monoenergetic depth doses to predict the depth dose distribution. Significance. The depth doses of the polyenergetic ion beams were predicted with mean relative errors of 2.20% and 1.95% for 11C and 15O ion beams, respectively. These errors represent the mean value of the differences between the predicted and measured values, normalized to the maximum measured value. The predicted distal fall-off positions at 80% of the Bragg peak were within 0.13 mm of the measured values for both ion beams. The applicability of the method was demonstrated for a head phantom irradiated with 11C ion beams using Monte Carlo simulation and the predicted distal fall-off positions at 80% of the Bragg peak was within 0.08 mm of the simulated values.