The following data taken from Nowak's classic paper shows the difficulty of relying on size to identify fossils.
Summary upper carnassial tooth length of canids: LP4
Canis dirus: 28.7 - 35.5
Canis lupus 22.2 - 30.5
Canis latrans 17.6 - 22.8
Canis familiaris 14.4 - 22.7
Canis armbrusteri 26.6 - 29.5
Canis edwardi 24.0
Canis lepophagus 19.0 - 20.7
In some cases (the fossils) the sample is small. In others, (lupus and latrans) it is more than 100 skulls.
Coyotes overlap with wolves, and wolves overlap with Dire Wolves, but in each case, a significant portion of the individuals can be allocated to the correct taxon by size alone. Domestic dogs are a mess, as they overlap with both coyotes and wolves. When you throw in the remaining fossil canids, size ends up being pretty useless as the sole determiner of identity. Usually some knowledge of the geologic context can help eliminate some of the fossil taxa.
This is an old data set, and I'll run a similar analysis of Tedford and Wang's data and post that later.