To check the association between a good performance of language and the recognition of facial emotional expressions in elderly individuals. Transversal study performed with 118 elderly individuals from the primary care services of health of a city in the state of São Paulo. Sociodemographic data were collected, regarding the performance of language through the domain of Addenbrooke Cognitive Examination - Revised and Recognition of Facial Emotional Expressions. The sample was divided in thirds according to the performance of language: T1 = the best, T2 = average, and T3 = the worst. The groups T1xT3 were compared regarding the performance of recognition of facial expressions of anger, disgust, fear, happiness, sadness, and surprise, and for the intensities of 40%, 60%, 80%, and 100%. The association of independent variables over the performance of language was analyzed through logistic regression. The multivariate model was built from the results of the univariate analyses and has included the continuous variables by emotion and by intensity. Age and schooling associated to the performance of language in the univariate model were included in the multivariate model in order to adjust association analyses. The sample was mainly female (84.7%), with an average age of 70.5 years old, and 3.5 schooling years. The variables associated to the best performance of language in comparative analysis of T1 and T3 were: surprise (OR = 1.485, IC 95% 1.194 - 1.846), and disgust (OR = 1.143, IC 95% 1.005 - 1.300). The recognition of facial emotional expressions of surprise and disgust were shown as important factors associated to the good performance of language.