Background: We evaluated the accuracy of code strokes activations at two comprehensive stroke centres in Toronto, Canada. Methods: We conducted a multi-centre, retrospective cohort study of all adult patients seen as code stroke in emergency rooms (ER) of two comprehensive stroke centres (CSC) in Toronto, Canada between January 1, 2022 and Dec 31, 2022. We included cases where the code stroke was activated in the field by paramedics and where it was activated in the ER by a physician. We reported off-criteria code stroke activations as the proportion of code stroke activations that did not meet all criteria for activation, and described the criteria that were not met. Results: A total of 677 (61.9% paramedic) code strokes were seen at CSC1 and 439 (80.6% paramedic) at CSC2. At CSC1, 21.2% paramedic-activated and 38.6% ER-activated were off-criteria, and at CSC2, 14.2% paramedic-activated and 48.1% ER-activated code stroke were off-criteria. Most of these were due to incorrect assessment of the last seen normal time. Conclusions: One in five code strokes did not meet criteria for activation. Improving the accuracy of paramedic and ER assessment of last seen normal time may be an avenue to reduce off-criteria code stroke activations.