Statistical matching tackles the problem of drawing information on a pair of random variables (Y, Z) which have not been observed jointly in one sample survey. In fact, Z andY are available in two distinct and independent surveys whose sets of observed units are non overlapping. The two surveys observe also some common variables X. This problem has traditionally been analyzed when the two sample surveys consists of independent and identically distributed observations from the same model. On the contrary most surveys, especially those managed in National Statistical Institutes, consists of samples drawn from a finite population according to complex survey designs. This paper compares some of the procedures described in the literature and analyzes their effects on the analysis of uncertainty, ie of the lack of joint information on the random variables of interest.