This paper suggests a new approach to constructing reliable distributed systems. The basic idea is to use a hierarchical framework as a reference model by which distributed systems can be constructed to be more reliable. Reliability is deemed to be a fundamental potential advantage of distributed systems. Existing research tends to concentrate on improving the reliability of specific system functions such as distributed file systems, distributed directory systems, and distributed transaction systems. This research usually focuses on one of the specific reliability issues in a system, as a result, the inter-relationships of these system functions are ignored. We propose a layer-structured framework which defines various reliability issues in distributed systems and relates these issues in a hierarchical way. The framework consists of four layers. Each layer uses the services provided by the lower layer and it in turn provides support for the upper layer. Lower layers provide failure abstraction to the upper layers. Functions at the upper layers can therefore be built on top of the functions provided by the lower layers without concern for failure handling. The advantages of our framework are integrated fault tolerance, elimination of redundancy, optimization of reliability, ease of evaluation, good modularity, and better code reusability.