The rate of sudden infant death syndrome (SIDS) declined significantly in Canada and the US between the late 1980s and the early 2000s. In the US, this decline was shown to be due in part to a shift in diagnosis, as deaths from accidental suffocation and strangulation in bed and from other ill-defined and unspecified cause increased concurrently. This study was undertaken to determine whether there was such a shift in diagnosis from SIDS to other causes of death in Canada, and to quantify the true temporal decrease in SIDS. Cause-specific infant death rates were compared across three periods: 1991–95, 1996–2000 and 2001–05 using the Canadian linked livebirth-infant death file. The temporal decline in SIDS was estimated after adjustment for maternal and infant characteristics such as maternal age and small-for-gestational age using logistic regression. Deaths from SIDS decreased from 78.4 [95% confidence interval (CI) 73.4, 83.4] per 100 000 livebirths in 1991–95, to 48.5 [95% CI 44.3, 52.7] in 1996–2000 and to 34.6 [95% CI 31.0, 38.3] in 2001–05. Mortality rates from other ill-defined and unspecified causes and accidental suffocation and strangulation in bed remained stable. The temporal decline in SIDS between 1991–95 and 2001–05 did not change substantially after adjustment for maternal and infant factors. It is unlikely that the temporal decline of SIDS in Canada was due to changes in cause-of-death assignment practices or in maternal and infant characteristics.