Methods and system perform tool tracking during minimally invasive robotic surgery. Tool states are determined. using triangulation techniques or a Bayesian filter from either or both non-endoscopically derived and endoscopically derived tool state information, or from either or both non-visually derived and visually derived tool state information. The non-endoscopically derived tool state information is derived from sensor data provided either by sensors associated with a mechanism for manipulating the tool, or sensors capable of detecting identifiable signals emanating or reflecting from the tool and indicative of its position, of external cameras viewing an end of the tool extending out of the body. The endoscopically derived tool state information is derived from image data provided by an endoscope inserted in the body so as to view the tool.