Mathematical evaluation techniques to derive the calibrated power spectrum of reflected ultrasonic echoes from ocular tumors can be used with a clinical computer system to objectively classify malignant melanoma and metastatic carcinoma of the eye. Sets of spectral data provide a reproducible standard for classification that closely approximates the Callender system used in histologic classification. We used spectral data with a computer data library to separate spindle B malignant melanomas (20 tests in 17 patients) from mixed-epithelioid melanomas (17 tests in 13 patients) and to identify metastatic carcinoma (20 tests in 17 patients) with approximately 98% reliability. Tissue structures can be acoustically stained in B-mode images to define the specific anatomic and structural properties that provide the acoustic differentiation. These data are obtained under in vivo conditions and allow a noninvasive classification of tumors in a way not previously possible, aiding in the definitive diagnosis of ocular tumors, as well as in the planning and monitoring of treatment.