This paper presents an algorithm for resolving pronominal anaphora, specifically identifying the noun phrase antecedents of third person pronouns and lexical anaphors (reflexives and reciprocals). The algorithm is implemented in Prolog and applies to the syntactic structures generated by McCord's Slot Grammar parser. It uses salience measures derived from syntactic structure and a dynamic model of attentional state to select the antecedent noun phrase (NP) of a pronoun from a list of candidates. The algorithm was tested on computer manual texts and a blind test with 360 pronoun occurrences, achieving 86% accuracy. An enhancement was introduced that statistically models semantic and real-world relations, but only marginally improved performance by 2%. The algorithm was compared with Hobbs' algorithm, which achieved 87% accuracy for intersentential anaphora, but RAP achieved 89% for intrasentential anaphora, resulting in a 4% overall success rate advantage. The algorithm also compares with other approaches, including the centering approach and models that use informational factors for antecedent ranking. RAP's success is attributed to its syntactic and morphological filters, salience weighting based on grammatical roles and proximity, and dynamic salience degradation. The algorithm's performance was further evaluated in a blind test with RAPSTAT, a system that uses statistically measured lexical preferences. RAPSTAT showed improved performance when combining salience and statistical data. The paper discusses the relative contributions of the algorithm's components, including the syntactic filter, salience weighting, and equivalence classes. The algorithm's effectiveness is attributed to its ability to handle complex cases, including pleonastic pronouns and intersentential anaphora, and its use of a dynamic model of attentional state. The results indicate that RAP's salience-based approach is more effective than Hobbs' search procedure for the text domain tested.This paper presents an algorithm for resolving pronominal anaphora, specifically identifying the noun phrase antecedents of third person pronouns and lexical anaphors (reflexives and reciprocals). The algorithm is implemented in Prolog and applies to the syntactic structures generated by McCord's Slot Grammar parser. It uses salience measures derived from syntactic structure and a dynamic model of attentional state to select the antecedent noun phrase (NP) of a pronoun from a list of candidates. The algorithm was tested on computer manual texts and a blind test with 360 pronoun occurrences, achieving 86% accuracy. An enhancement was introduced that statistically models semantic and real-world relations, but only marginally improved performance by 2%. The algorithm was compared with Hobbs' algorithm, which achieved 87% accuracy for intersentential anaphora, but RAP achieved 89% for intrasentential anaphora, resulting in a 4% overall success rate advantage. The algorithm also compares with other approaches, including the centering approach and models that use informational factors for antecedent ranking. RAP's success is attributed to its syntactic and morphological filters, salience weighting based on grammatical roles and proximity, and dynamic salience degradation. The algorithm's performance was further evaluated in a blind test with RAPSTAT, a system that uses statistically measured lexical preferences. RAPSTAT showed improved performance when combining salience and statistical data. The paper discusses the relative contributions of the algorithm's components, including the syntactic filter, salience weighting, and equivalence classes. The algorithm's effectiveness is attributed to its ability to handle complex cases, including pleonastic pronouns and intersentential anaphora, and its use of a dynamic model of attentional state. The results indicate that RAP's salience-based approach is more effective than Hobbs' search procedure for the text domain tested.