Abstract This paper introduces a new framework for polygonal data analysis in the symbolic data analysis paradigm. We show that polygonal data generalizes bivariate interval data. A way for aggregating data in classes is presented to obtain symbolic datasets and, descriptive statistics (for instance, mean, variance, covariance, and histogram) and a linear regression model are proposed for symbolic polygonal data. A simulation study to available the performance of the polygonal linear regression based on a mean square error of area is done. The proposed methodology is applied to two real symbolic datasets represented by classes, and the results illustrate the usefulness of the statistical techniques.