With the development of digital communication systems, channel estimation becomes crucial for reliable signal transmission. Conventional channel estimation methods are easy to implement and work well for time-invariant or slow time varying channels; however, they cannot provide satisfactory results when channel characteristics change fast in wireless communication systems. This paper investigates the application of artificial neural networks (ANNs) to improve existing channel estimation methods. A two stage, hybrid approach that combines conventional least squares (LS) algorithm and machine learning techniques are proposed and tested on Rayleigh fading channels with different signal-to-noise ratios (SNRs). Both multi-layer feed forward neural networks (FNNs) and convolutional neural networks (CNNs) are considered. Computer simulation results show the proposed LS-CNN approach can significantly improve channel estimation and reduce the signal BER (bit error ratio) in wireless communication systems.