Oscillations and rhythms are measured in the brain through large-scale measures like EEG (electroencephalogram) and LFP (Local Field Potential). Particularly, cortical gamma rhythms (30-90 Hz) found in different brain regions are correlated with different cognitive states. Despite vast differences in the range frequencies in gamma rhythms, the regions communicate to complete high-level tasks. One way in which this takes place is entrainment, where the postsynaptic group phase-lock to the rhythmic input from the presynaptic group (constant phase-shift). Mathematical models of the neurons and the neural networks are proposed to uncover the mechanisms behind experimentally observed phenomena. Most works have used homogeneous models of spiking networks. These simplified models provide a valuable understanding of neural dynamics. However, neural heterogeneity (variation in the neural or network parameters) has been experimentally observed and shown to have a non-trivial effect on many neural processes. Few studies have dealt with the role of different types of neural heterogeneity in the entrainment of a large network, and how it affects the frequency range the neural network entrains to. In this project, we aimed to show how different types of network heterogeneity affect the ability of the networks to entrain to gamma frequencies. We used the Pyramidal-Interneuronal Network Gamma (PING) model, a model consisting of excitatory pyramidal cells (E-cells) and inhibitory interneurons (I-cells) that are synaptically connected and generate gamma oscillations. We show that heterogeneity in the synaptic conductance from excitatory neurons to inhibitory neurons greatly increases the frequency range over which the network can entrain. The mechanism that allows this to happen requires the heterogeneity to 1. Create an I-cell excitability gradient; 2. Introduce input synchrony difference among the I-cells. The entrained I-cell subsets formed under these two conditions are necessary for well-enhanced entrainment as they support the entrainment of the whole network through feedback inhibition. This improvement is shown to be robust in large parameter space.