The common construction takes the average of the largest annual drawdowns over the period - often three years - and divides annualised return by that average, sometimes with a fixed 10% added to the denominator in the original formulation.
The point of averaging is statistical. max-drawdown is one draw from a distribution and moves wildly between samples; the mean of the three worst is a more stable estimate of what a bad year actually looks like. That makes Sterling better for comparing strategies and worse for answering how bad it can get.
Definitions vary between providers - some omit the 10% constant, some use a different count of drawdowns - so never compare Sterling ratios from two sources without checking the formula. That ambiguity is the main reason it is less quoted than calmar-ratio despite being the better-behaved measure.
Related: calmar-ratio, mar-ratio, max-drawdown, ulcer-index