Passive cavitation images (PCIs) generated from scattered acoustic waves are a potential technique for monitoring lesion formation during high-intensity focused ultrasound (HIFU) thermal ablation. HIFU lesion prediction by PCIs was assessed in ex vivo bovine liver samples (N = 14) during 30-s sonications with 1.1-MHz continuous-wave ultrasound (1989 W/cm2 estimated spatial-peak intensity). Treated samples were sectioned, optically scanned, and the HIFU lesions segmented based on tissue discoloration. During each insonation, a 192-element, 7-MHz linear array (L7/Iris 2, Ardent Sound) passively recorded emissions from a plane containing the HIFU propagation axis oriented parallel to the image azimuth direction. PCIs were formed from beamformed A-lines filtered into fundamental, harmonic, ultraharmonic, and inharmonic frequency bands. Lesion prediction was tested using binary classification of local tissue ablation based on thresholded PCIs, with spatial specificity and sensitivity of lesion prediction quantified by the area under receiver operating characteristic curves (AUROC). Tadpole-shaped lesions were best predicted by harmonic emissions (AUROC = 0.76), prefocal lesions were best predicted by harmonic or ultraharmonic emissions (AUROC = 0.86), and cigar-type focal lesions were best predicted by fundamental and harmonic emissions (AUROC = 0.65). These results demonstrate spatial specificity and sensitivity when predicting HIFU lesions with PCIs. [Work supported in part by NIH grants F32HL104916 and R21EB008483.]