Wireless Sensor Networks are moving out of the laboratory and into the field. For a number of reasons there is often a need to update sensor node software, or node configuration, after deployment. The need for over-the-air updates is driven both by the scale of deployments, and by the remoteness and inaccessibility of sensor nodes. This need has been recognized since the early days of sensor networks, and research results from the related areas of mobile networking and distributed systems have been applied to this area. In order to avoid any manual intervention, the update process needs to be autonomous. This paper presents a comprehensive survey of software updating in Wireless Sensor Networks, and analyses the features required to make these updates autonomous. A new taxonomy of software update features and a new model for fault detection and recovery are presented. The paper concludes by identifying the lacunae relating to autonomous software updates, providing direction for future research.