This study compared 3 systems and a newly designed stepwise discriminant diagnostic system (SDDS) to assess accuracy, reproducibility, and reliability in adrenocortical nodular hyperplasia (ACNH; n = 82), adenoma (ACA; n = 78), and carcinoma (ACC; n = 32) (diagnoses according to World Health Organization criteria; median follow-up, 135 months). In cross-validations, we studied cortex appearance, growth pattern, cytoplasmic features, nuclear parameters, mitotic figure counting (MFC), necrosis, invasion, and stromal-inflammatory reactions. The SDDS independent predictors were MFC/high-power field SD, anisokaryosis, cortex appearance, and stromal reaction, correctly classifying 100% of ACNH, 91% of ACA, and 88% of ACC cases. The Hough system correctly classified 78% of ACNH, 81% of ACA, and 84% of ACC cases; the Weiss and van Slooten systems correctly classified 100% of ACNH, 0% of ACA, and 92% of ACC cases. MFC variability is the most important adrenocortical malignancy criterion. Accurate malignancy diagnosis requires multiple variable evaluations, provided by SDDS (the most specific system) and the Weiss or van Slooten system (the most sensitive). SDDS is the most useful system for distinguishing tumors from ACNH (myxoid stroma and anisokaryosis).