The negative binomial distribution models the number of failures before achieving a specified number of successes in a sequence of independent and identically distributed Bernoulli trials. It generalizes the geometric distribution and is particularly useful for overdispersed count data where the variance exceeds the mean.