Recognition of the potential benefits of being able to variably apply herbicides based on the sensing of weeds has led to much research and development activity. The purpose of this paper is to survey what work has been done already with a view of formulating future research and development directions with a goal of the development of practical weed sensing technology. Two approaches have typically been used for weed detection. The first is the photo-detection approach, which measures the average reflected light from the field of view of the detector. Light-detecting (photo) diodes or resistors have been used in this low resolution approach meaning that the area sensed by one sensor is large. The other approach is the machine vision approach. In this approach, digital images of the field scene are acquired with some type of camera, and the information contained in these images is processed by a computer to retrieve knowledge or understanding of the scene. Examples of both of these approaches will be discussed in this paper.