||A system is developed to automatically generate control functions for linked mechanisms that express specific emotions. This system could be employed to produce motions for characters in a computer animation. It takes advantage of the fact that evaluation functions, which take physical characteristics of a motion sequence as input, can be used as a filter on a set of motion sequences to choose one which expresses a desired emotion. In other words, the evaluation function is able to quantify how much of a certain emotion is being displayed by a motion sequence.; Thus, the system hinges on finding an evaluation function that determines how much of a certain emotion is being displayed based on a given set of physical characteristics. This is accomplished through the following steps: ; 1. Motion sequences are generated for each geometry. ; 2. Physical characteristics are measured for each motion sequence. ; 3. Humans rank each motion sequence in four emotion categories through survey.; 4. A mapping from the physical measures to the emotional categories is deter; mined. ; 5. This mapping is used as a filter in the selection of control functions.; The control functions, for the purposes of this research, are represented as sensor-actuator networks (SANs). A control function that produces motion with a desired emotional content can be created automatically by generating many random SANs for a given mechanism geometry, and then using an evaluation function that recognizes the desired emotion as a filter in order to select the SAN whose motion; scores the highest.