J. C. Gower introduced a general coefficient of similarity to measure the similarity between two sampling units. The matrix of similarities between all pairs of sample units is shown to be positive semidefinite (except possibly when there are missing values). This property is important for multidimensional Euclidean representation of the sample and establishes inequalities among similarities relating three individuals. The coefficient is extended to handle a hierarchy of characters. The coefficient measures resemblance between individuals based on presence/absence or observed values of qualitative or quantitative properties. For dichotomous characters, presence is denoted by + and absence by -; for qualitative characters, similarity is 1 if individuals agree and 0 if they differ; for quantitative characters, similarity is calculated based on the range of values. The similarity between two individuals is the average score over all possible comparisons. The coefficient is related to other similarity coefficients, such as Sneath's and the simple matching coefficient. The positive semidefinite property of the similarity matrix allows for Euclidean representation and ensures the triangle inequality holds. The matrix is also p.s.d. when weights are applied, provided weights are non-negative. Weighting and hierarchical characters are discussed, with the need to ensure that primary characters are not overshadowed by secondary characters. The coefficient can be adjusted to handle hierarchical data, ensuring that primary character matches are prioritized. The coefficient is flexible and can handle various data types without reprogramming. The coefficient has been used in hierarchical cluster analysis and principal coordinate analysis. It is flexible and suitable for multistate and quantitative characters. The p.s.d. property is important for numerical methods and interpretation of cluster and ordination analyses. The coefficient is also suitable for hierarchical data, though its use is not always necessary. The coefficient is p.s.d. when all characters are qualitative, quantitative, or dichotomous. It remains p.s.d. for any combination of these types. Missing values can cause the matrix to lose its p.s.d. property, but this can be mitigated by replacing missing values with appropriate values. The coefficient is flexible and can be used in various contexts, including taxonomy and classification.J. C. Gower introduced a general coefficient of similarity to measure the similarity between two sampling units. The matrix of similarities between all pairs of sample units is shown to be positive semidefinite (except possibly when there are missing values). This property is important for multidimensional Euclidean representation of the sample and establishes inequalities among similarities relating three individuals. The coefficient is extended to handle a hierarchy of characters. The coefficient measures resemblance between individuals based on presence/absence or observed values of qualitative or quantitative properties. For dichotomous characters, presence is denoted by + and absence by -; for qualitative characters, similarity is 1 if individuals agree and 0 if they differ; for quantitative characters, similarity is calculated based on the range of values. The similarity between two individuals is the average score over all possible comparisons. The coefficient is related to other similarity coefficients, such as Sneath's and the simple matching coefficient. The positive semidefinite property of the similarity matrix allows for Euclidean representation and ensures the triangle inequality holds. The matrix is also p.s.d. when weights are applied, provided weights are non-negative. Weighting and hierarchical characters are discussed, with the need to ensure that primary characters are not overshadowed by secondary characters. The coefficient can be adjusted to handle hierarchical data, ensuring that primary character matches are prioritized. The coefficient is flexible and can handle various data types without reprogramming. The coefficient has been used in hierarchical cluster analysis and principal coordinate analysis. It is flexible and suitable for multistate and quantitative characters. The p.s.d. property is important for numerical methods and interpretation of cluster and ordination analyses. The coefficient is also suitable for hierarchical data, though its use is not always necessary. The coefficient is p.s.d. when all characters are qualitative, quantitative, or dichotomous. It remains p.s.d. for any combination of these types. Missing values can cause the matrix to lose its p.s.d. property, but this can be mitigated by replacing missing values with appropriate values. The coefficient is flexible and can be used in various contexts, including taxonomy and classification.