It is not entirely obvious what is meant by the average (or mean) value of a function on an interval. We know how to find the mean of a finite collection of numbers (their sum divided by their number). Needless to say, we run into problems when we want to talk about the mean of all the values of a function on a particular interval, since they are infinite in number.

To find our way out of this conundrum, we recall the definition of the n-th (upper) Riemann sum for the function f on the interval [a, b]:

 Un(f, a, b) =  Mi

Note that Un(f, a, b) is equal to to the product of b - a (the length of the interval) and the mean of the values of f at n more or less evenly-spaced points in the interval. Clearly this is a reasonable approximate mean of the function f on the interval [a, b].

Naturally, the same is true for the nth lower Riemann sum. As n gets larger and larger, we might imagine the upper and lower Riemann sums to approach (one from above, one from below) the product of b - a and some "true" mean of the function f on [a, b]. Indeed, this indicates precisely how we will define the average value, denoted . We set =  Un(f, a, b) =  Ln(f, a, b) =  f (x)dx

There is a way of seeing graphically that this definition makes sense. An easy computation shows that the integral of the constant from a to b is equal to that of the function f (x):  dx = |ab = (b - a) = f (x)dx

Thus, is the height of a rectangle of length b - a that will have the same area as the region below the graph of f (x) from a to b. In physical terms, if f (t) represents the velocity of a moving object, then another object moving with velocity will travel the same distance between the moments t = a and t = b.