Natural forest comprises approximately 78% of New Zealand’s total forest area and their dead wood carbon pools contribute to the national carbon balance reported under the United Nations Framework Convention on Climate Change. We investigate the accuracy of coarse woody debris (CWD) carbon stock estimates obtained from periodic inventories of New Zealand’s natural forest. Dimensions and decay classes of standing-dead spars, fallen logs and stumps over 10 cm diameter were assessed in permanent plots during two inventory cycles (2002–07 and 2009–14). Carbon in CWD was estimated by multiplying CWD volume with species-specific mean wood and bark density, a density modifier based on decay class, and the carbon fraction. The measured change in CWD between inventories was obtained by subtracting 2002–07 from 2009–14 mean CWD carbon stocks. This measured change in CWD carbon stocks was compared with the modelled change derived by applying a decay model to 2002–07 stocks to determine the losses from decay and adding inputs from the tree mortality recorded between the two measurement cycles. Estimates of CWD carbon stocks from field measurements were 19.9 and 19.5 tC ha−1 in 2002–07 and 2009–14 respectively, with a statistically non-significant change of −0.4 ± 0.6 tC ha−1 (95% C.I.) over the period between measurements implying that CWD carbon is in a steady state. However, the modelling approach gave a significant increase of 2.5 ± 0.7 tC ha−1, implying that CWD stocks are increasing. It appears likely that the primary reason for the discrepancy between the direct measurement and modelling approaches is that CWD carbon stocks are substantially underestimated using current measurement protocols used in the inventory of New Zealand’s natural forest. If CWD carbon stocks in New Zealand’s natural forests are approximately in balance as implied by the direct measurement of CWD stocks, CWD inputs from mortality would have to be balanced by losses from decay. For this to be true, the CWD pool would have to approximate 35.4 tC ha−1 which is 78% higher than its measured value.