In Monte Carlo simulation of CT dose, many input parameters are required (e.g. bowtie filter properties and scan start/end location). Our goal was to examine the uncertainties in patient dose when input parameters were inaccurate. Using a validated Monte Carlo program, organ dose from a chest CT scan was simulated for an average-size female phantom using a reference set of input parameter values (treated as the truth). Additional simulations were performed in which errors were purposely introduced into the input parameter values. The effects on four dose quantities were analyzed: organ dose (mGy/mAs), effective dose (mSv/mAs), CTDIvol-normalized organ dose (unitless), and DLP-normalized effective dose (mSv/mGy · cm). At 120 kVp, when spectral half value layer deviated from its true value by ±1.0 mm Al, the four dose quantities had errors of 18%, 7%, 14% and 2%, respectively. None of the dose quantities were affected significantly by errors in photon path length through the graphite section of the bowtie filter; path length error as large as 5 mm produced dose errors of ⩽2%. In contrast, error of this magnitude in the aluminum section produced dose errors of ⩽14%. At a total collimation of 38.4 mm, when radiation beam width deviated from its true value by ± 3 mm, dose errors were ⩽7%. Errors in tube starting angle had little impact on effective dose (errors ⩽ 1%); however, they produced organ dose errors as high as 66%. When the assumed scan length was longer by 4 cm than the truth, organ dose errors were up to 137%. The corresponding error was 24% for effective dose, but only 3% for DLP-normalized effective dose. Lastly, when the scan isocenter deviated from the patient’s anatomical center by 5 cm, organ and effective dose errors were up 18% and 8%, respectively.