Mixing time of markov chain

A random walk on a graph can be treated as a Markov Chain. Several
references say that mixing time of Markov-chain on a graph of n
vertices is of the order of

log(n)/(spectral gap)

where spectral gap is the difference between top 2 eigenvalues of the
transition probability matrix of the Markov chain.  I am yet to find a
formal proof of this statement ! Can anybody tell  about a reference in
web that shows the proof?