Machine learning models are capable of estimating strength parameters in concrete with high accuracy. For any of these, an initial exploratory analysis of the dataset is highly recommended, with investigation of their influence on the property or characteristic to be estimated, as well as their correlations. This study presents this analysis for a set of 505 concrete mixtures, containing consumption of raw materials, additives and additions, including specific masses of the basic material. As a result, in addition to the expected basic relationships, others not so explicit were evidenced, such as the strong influence of some specific masses on the compressive strength. Given the strong relationship of the characterization parameters, its need for the construction of more refined models, with application in different sources of inputs, is evidenced.