A model of a (convex) technology of representative and non-representative firms in a heterogeneous sector is presented in non-parametric and parametric versions. The heterogeneity is specified with error terms. The models including a non-parametric distribution of the errors can be estimated with entropy econometrics from firm survey data. This requires two important modifications in the standard approach to entropy estimation of Golan, Judge and Miller: The compact support of the probability distribution should be designed to capture eventual non-zero covariance. And cross-entropy need to be redefined for cases of multiple observations.