We study how individuals repay their debt using linked data on multiple credit cards from five major issuers. We find that individuals do not allocate repayments to the higher interest rate card, which would minimize the cost of borrowing. Instead, individuals seem to allocate repayments using a balancing-matching heuristic by which the share of repayments on each card is matched to the share of balances on each card. We show that balance matching captures more than half of the predictable variation in repayments, performs substantially better than other models, and is highly persistent within individuals over time. Consistent with these findings, we show that machine learning algorithms attribute the greatest variable importance to balances and the least variable importance to interest rates in predicting repayment behavior.