Cache-enabled small cell networks have been regarded as a promising approach for network operators to cope with the explosive data traffic growth in future 5 G networks. However, the user association and resource allocation mechanism has not been thoroughly studied under given content placement situation. In this paper, we formulate the joint optimization problem of user association and resource allocation as a mixed integer nonlinear programming(MINLP) problem aiming at deriving a balance between the total utility of data rates and the total data rates retrieved from caches. To solve this problem, we propose a distributed relaxing-rounding method. Simulation results demonstrate that the distributed relaxing-rounding method outperforms traditional max-SINR method and range-expansion method in terms of both total utility of data rates and total data rates retrieved from caches in practical scenarios. In addition, effects of storage and backhaul capacities on the performance are also studied. Cache-enabled small cell networks have been viewed as a promising approach for network operators to cope with the explosive data traffic growth in future 5 G networks. However, the user association and resource allocation mechanism has not been thoroughly studied under given content placement situation. In this paper, we formulate the joint optimization problem of user association and resource allocation as a mixed integer nonlinear programming (MINLP) problem aiming at deriving a balance between the total utility of data rates and the total data rates retrieved from caches. problem, we propose a distributed relaxing-rounding method. simulation results demonstrate that the distributed relaxing-rounding method outperforms traditional max-SINR method and range-expansion method in terms of both total utility of data rates and total data rates retrieved from caches in practical In addition, effects of storage and backhaul capacities on the performance are also stu died.