Central banks manage about \$12 trillion in foreign exchange reserves, influencing global exchange rates and asset prices. However, some of the largest holders of reserves report minimal information about their currency composition, hindering empirical analysis. I describe a Hidden Markov Model to estimate the composition of a central bank's reserves by relating the fluctuation in the portfolio's valuation to the exchange rates of major reserve currencies. I apply the model to China and Singapore, two countries that collectively hold about \$3.4 trillion in reserves and conceal their composition. I find that both China's reserve composition likely resembles the global average, while Singapore probably holds fewer US dollars.