By definition, portfolio drawdown is a drop in the portfolio value compared to the previous maximum. We study a measure of risk, which depends on the portfolio drawdown curve (also called underwater curve) considered in active portfolio management. The new risk measure, Conditional Drawdown-at-Risk (CDaR), is defined as the mean of the worst x% drawdowns. This measure of risk is closely related to the Conditional Values-at-Risk risk measure. The CDaR risk measure has several important properties, which make it attractive from a practical perspective: (1) compared to variance or Value-at-Risk (VaR), it adequately reflects investors' preferences; (2) it is robust: it depends upon many significant drops in the portfolio value rather than on one extreme event; (3) information on sequence of evens is not lost (compared to approaches such as VaR or variance); (4) minimal data requirements: historical data can be directly used for path generation; (5) the technique is very stable numerically; (6) can be efficiently implemented using Linear Programming techniques. Some practical recommendations on how to use the CDaR measure for getting practically stable portfolios are provided. Using CDaR, we solved a real life allocation problem for a portfolio of derivatives.