A comprehensive understanding on the dynamics of droplet clustering in sprays is important for a wide range of applications ranging from combustion engines to spray drying. Due to inherent flow inhomogeneity and air entrainment, the mechanism of dispersion of droplets in sprays is not the same as that in homogeneous and isotropic turbulence, which has been relatively well studied. The current research work aims to understand the physics of clustering of polydispersed water droplets in air-assist sprays. In particular, the focus is on studying the influence of local liquid mass loading on droplet clustering in the spray. For this purpose, the injector was operated for different inlet air and water flow rates. The droplet clusters as well as voids were uniquely characterized by the application of the Voronoi analysis on the spray images captured using the particle image velocimetry technique far downstream of the injector exit. The interferometric laser imaging for droplet sizing technique was used to measure the droplet size distribution. The droplet clustering in the spray was found to be a multi-scale process that spans over a wide range, from tens of the Kolmogorov scale up to a fraction of spray half-width. Similar normalized distributions of cluster and void areas for different cases suggested self-similarity of the clustering process. A new method for characterizing the length scale of droplet clusters was described. The average cluster length scale and the degree of clustering were found to strongly depend on local liquid mass fraction and gas phase Reynolds number. The radial evolution of droplet cluster statistics across the spray was also examined. © 2018 Author(s).