In recent years, several Artificial Neural Network (ANN) models have been developed as surrogate constitutive models for composites [1]. These ANN models are developed based on two main categories of data, namely experimental results and physics-based simulations. However, one of the major challenges for developing deep learning models using ANNs is still a requirement for a huge amount of data for training and validation purposes. Acquiring a substantial amount of data is sometimes very challenging, particularly when using high-fidelity full-field simulations. This study examines the potentials of a new ANN approach, referred to as Generative Adversarial Networks (GANs), for data generation in material science. Recently introduced GANs, initially proposed by Goodfellow et al. [2], are a class of artificial neural networks that are used for generating synthetic data. They consist of two main components: a generator and a discriminator. The generator creates synthetic data samples, while the discriminator attempts to distinguish the synthetic samples from real samples. The two networks are trained together in an adversarial manner, with the goal of the generator producing synthetic samples that are indistinguishable from real ones. In many applications, such as computer vision, speech synthesis, and natural language processing, GANs have been used to generate a variety of synthetic data including images, videos, and audio. The focus of this work is on micro-mechanical elastic properties of short fibre reinforced composites (SFRCs). Mentges et al. [3] developed a micromechanics-based ANN model for the elastic properties of SFRCs. Matrix and fibre elastic properties, fibre volume fraction, fibre orientation distribution, and fibre geometry are inputs to the ANN model, and the stiffness tensor is the model output. A comprehensive data set was developed using a Finite-Element-based orientation averaging method. In this work, we used GANs with a part of the data set from [3] to generate synthetic data representing the existing pattern in the real data set. According to the obtained results, the generated data exhibits similar patterns to the original data.