The above post gives the method to be followed. I wand to add some points to that.
Out of the two (or four if differential) input terminals, you give a fixed voltage (say Vr) to one of the terminal (say V+). This will work as reference to the comparator. Now apply a square wave to the other terminal (say V-).
-->If the value of Vr is some where between the ICMR of the comparator (excluding the extreme points), the comparator output sould change. Now, you have two points in the transient analysis (a) the time when the square wave (V-) crosses Vr and (b) when the comp. output changes its decission. The time difference between (a) and (b) should be the propagation delay.
One important point must be kept in mind that the voltage levels of the V+ and V- are important. Because, to detect little voltage difference the comp. should take more time than it should take to detect higher voltage difference at V+ and V-. Keep this in mind. In my opinion instead a generic term "propagation delay", it is better to say "minimum propagation delay" or "maximu propagation delay".