Progressive penalties are common in real-world deterrence systems, yet the evidence on fines mainly concerns raising their level rather than steepening their gradient. We compare progressive and linear fine schedules in a laboratory tax-evasion game with complementary between- and within-subject designs. The implemented progressive schedule increases total revenue by about 9% between subjects and, under additivity, about 7% per switch within. Between subjects, average evasion does not change detectably. Even the largest evasion response our equivalence analysis permits would move expected revenue by less than 40% of the unadjusted between-subject difference. Because most evaded income falls in the highest-penalty brackets, the schedule also raises the evasion-weighted fine rate, which accounts for much of the aggregate gain. Once the level difference is rescaled away, curvature alone shifts revenue from smaller to larger offenses at fixed behavior. Revenue also rises among subjects predicted to evade most, the only ones for which our model signs the effect on total revenue. Overall, the schedule we implement raises and reallocates revenue more clearly than it changes evasion. Policymakers seeking to reduce evasion should pair such reforms with complementary measures that the additional revenue can help finance.