Attributes are Chef's way of storing configuration data and can be thought of as a large, but disjointed, hash structure. Chef pulls data from various locations and combines that data in a specific order to produce the final hash of attributes. This data is computed when a client requests its run list from the server (such as when you execute chef-client on a node). This mechanism allows you to describe data with a higher level of specificity at each step of the process, decreasing in scope going from the cookbook attributes files down to node-specific configuration data.
For example, imagine you are deploying PostgreSQL onto the hosts in your infrastructure. With PostgreSQL, there are a very large number of configuration options that can be tuned, ranging from open ports and number of concurrent connections down to memory used for key caches and other fine-grained configuration options. The cookbook's attributes files should provide enough configurations for PostgreSQL to work...