Introduction: diabetes is an eligible model of chronic disease for the study of Health Literacy (HL). Based on the theoretical model proposed by Sørensen and collaborators, the instrument “Health Literacy in Diabetes” (HLD) was developed in Brazil and contains ten questions. Regard to access, the HLD contemplates the number of health professionals who provide the information, the topics on the information considering themes that affect the course of the disease and the sources from which the information is obtained. These items were summarized in the first three questions of the questionnaire. The following seven questions were developed to measure the understanding, evaluation and application of the information. They ask about the classification of the information as more or less important, the identification of the quality and veracity of the information and the evaluation of the advantages and disadvantages of the different types of therapy. In addition, it asks whether the interviewee practices the information in his daily life, maintaining an appropriate behavior considering the information received. Objective: construct validation of the HLD combined with logistic regression to analyze the relationship between latent constructs, clinical and sociodemographic variables. Method: the instrument was applied among people with diabetes assisted in primary health care in a city in the southeastern region of Brazil (n=341). The items were grouped into three latent variables. The construct validity was estimated by Confirmatory Factor Analysis (CFA) and logistic regression was used to identify variables associated with HL. Results: greater access implied an increase in other HL skills. The CFA demonstrated adjustment and plausibility of the model with adequate values ​​of χ2 / dl, RMSEA, GFI, CFI and TLI. The model presented VME and CR with values ​​compatible with a good internal consistency. High levels of HL were associated with the male gender and with a higher level of education. There was no statistically significant difference in relation to the measurement of glycated hemoglobin. Conclusion: the modeling of structural equations allowed the analysis of associations between latent variables and sociodemographic and biochemical variables, which can be useful for the development of diabetes management strategies.