Copulas are an exremely useful tool used to build models of the joint behavior of multiple financial variables. They allow you to define a multivariate statistical distribution in two steps:
Basically, what the copula does is to specify the structure of the dependence among the variables, leaving their marginal distributions unaltered.
Mathematically, a copula is a scalar-valued function of n-variables. If you plug n univariate distribution functions into its n arguments you get a multivariate distribution function, which has the original n distribution functions as its marginals. Stated more formally (for the case of two variables), if C=C(u,v) is the copula and F(x) and G(y) are two univariate distribution functions, H=H(x,y)=C(F(x),G(y)) is a bivariate distribution function having C and G as its marginals.