In document analysis and recognition applications, segmenting the text lines accurately is a crucial step. Text line segmentation in handwritten documents is still a challenging task because of numerous factors that can decrease the segmentation accuracy. In this work, generative adversarial networks are proposed to segment the text lines in handwritten documents and text line segmentation problem is considered as an image-to-image translation problem and generative models are used to extract text lines from document images. Generative model is trained with a diverse and challenging Arabic dataset and segmentation performance of the method is evaluated with visual and numerical results. Proposed generative model can segment the text lines having 0.81 precision, recall and F-measure results. Also, visual results show that generative model is highly capable of segmenting the text lines having various behaviors.