With the spread of systems approaches to biological research, demand for methods and tools capable of extracting quantitative measurements from individual and timebased sequences of microscope images is increasing. In the plant sciences, the model plant Arabidopsis thaliana receives much attention, with confocal laser microscope images a dominant imaging modality. In confocal laser microscopy, images are formed by excitation of fluorescent proteins attached to features of interest. The data required from these images is diverse, and includes objects’ volumes, growth rates, lengths, fluorescent properties etc. Its extraction requires objects’ identity to be determined within an initial image, and subsequently maintained over time. Many of the images show multiple components, such as organised networks of cells, making high-level models of the structure of the viewed samples necessary. Confocal images of Arabidopsis roots show cells which are clustered together, connected by shared object boundaries, forming a network topology. We employ a variant of the recently developed network snakes approach to identify cell boundaries in confocal images of growing Arabidopsis roots. The original snake energy function employed is modified to reflect the task at hand. Network snakes both make explicit the connections between adjacent region boundaries (cell walls) and describe boundary shape. The network is initialised using a two-level variant of the watershed algorithm. Given image sequences showing slow-growing plants, the implicit tracking ability of the network snake can be exploited to maintain cell identity over time. In many cases, however, growth between frames is sufficient to make this impossible;  the snake network cannot reacquire the correct boundary set when reinitialised in the same configuration on the next image. This problem is addressed by incorporating a multi-target particle filter-based tracker. This tracks the locations of network nodes between frames, allowing the snake to be warped onto the new image before its energy function is minimised. The combined approach is demonstrated using time-varying confocal images of the elongation zone of the Arabidopsis root in which the plasma membrane is marked with a fluorophore. A software tool, CellSET, based on the methods outlined here has been developed, and used to provide quantitative input to mathematical modelling tools at the BBSRC/EPSRC Centre for Plant Integrative Biology.