Two-stage Stochastic Programming (TSP) was used to optimize the allocation of water resources between cities in Heilongjiang Province with different inflow and risk levels, and the problems of optimal allocation and uncertainty in water resources system were solved. Conditional value at risk (CVaR) was used to measure the risks in water resources system to avoid the risks and maximize the economic benefits. When risk preference is not taken into account, the total revenue of Heilongjiang under the low, medium and high scenarios are (524.535~793.332) billion yuan, (861.664~962.580) billion yuan and 1.53.050 billion yuan respectively. When the risk preference value is from 0.1 to 0.9, the amount of water allocated by each city’s tertiary industries decreases with the increase of risk preference, and the administrators can choose appropriate water allocation schemes according to the degree of risk aversion of different industries. This study provides references for scientific allocation of urban water resources management.