Using a unique institutional setting of dual exchange rates of Chinese currency, this paper provides novel evidence that firms manipulate trade data to evade capital controls. We develop a model showing that the trade data over-reporting is positively (negatively) correlated with the exchange rate spread when the spread is positive (negative), and such correlations are more pronounced for products with low risks of being caught. Empirical results from threshold regressions using time series data and Benford’s law using firm-product trade data between mainland China and Hong Kong support the theoretical predictions of dual exchange-rate arbitrage camouflaged under the trade account. Our results highlight the challenges to implementing capital controls and may help improve the effectiveness of such policies.