If we are just interested in the running time, we can think of MergeSort as operating like this:

Mergesort(input of size n)
{
Mergesort(input of size n/2)
Mergesort(input of size n/2)
Do O(n) work merging the two sorted lists
}

...where the "input of size n/2" is actually either floor(n/2) or ceiling(n/2). Since Big Oh doesn't care which, we will ignore the floors and ceilings and act as if the n is always a power of two.

In this case, it is easy to see that the recurrence relation

T(n) = 2T(n/2) + Theta(n)

describes the running time of the algorithm. You can now use the Master Theorem to analyze this recurrence.