Connections form between neurons during neural development guided by growth cones, highly dynamic structures at the tips of growing axons. Understanding the biological mechanisms underlying this guidance requires understanding the dynamic morphology of growth cones, and this requires the segmentation of growth cone outlines from potentially very long timelapse movies. Previous approaches to this problem have been based either on time-consuming human input, or on algorithms customised for specific datasets. Here we evaluate the effectiveness of deep convolutional neural networks (CNNs) for growth cone segmentation. First, we apply a deep CNN to a standard benchmark cell nuclei segmentation problem, and show that it achieves performance comparable to standard image processing techniques after training with only one image. Second, we show that the deep CNN can achieve good segmentation of a large set of phase-contrast timelapse movies of growth cones after training with only a few frames. Thus, deep CNNs provide a general method for growth cone image segmentation that is broadly applicable and require very little training.