This paper investigates the performance of graph colouring schemes for frequency assignment in Long-Term Evolution (LTE) networks with ultra-dense femtocells. The aim of the study is to minimize interference in such networks while ensuring efficient spectrum use for these femtocells. The three schemes investigated are the conventional greedy graph colouring algorithm, the saturation degree algorithm and our proposed graph-based theory (GBT) algorithm. The process of frequency assignment is similar in the last two except that the proposed GBT partitions the femtocells into independent sets for an efficient frequency re-use. The performance of these three schemes was analyzed through extensive simulations to determine the SINR and network capacity that can be obtained with the deployment of these schemes using the ITU-R P1238-7 path loss model. The outcome of this study showed that with the absence of a dynamic frequency assignment scheme, interference level is increased as the number of femtocell users within a particular coverage is increased, leading to a reduction in the capacity of such networks. Simulation results showed that all three algorithms considered have the ability to allocate frequencies to femtocells and minimize interference in a densely deployed environment, thereby increasing network capacity. The proposed GBT assigned the least sub-band thereby ensuring spectral efficiency while minimizing harmful interference. Results show that the greedy algorithm has a disadvantage of inefficiently assigning sub-bands randomly, while the saturation degree assigns more sub-bands when compared with the GBT scheme.Keywords — Femtocell, graph colouring, frequency assignment, LTE.