Complete data matrices are required for some statistical analysis techniques, making imputation of missing data necessary in certain circumstances. The Krzanowski imputation system is based on singular value decomposition of a matrix and has no distributional or structural assumptions, but the system needs an imputation refining process through an iterative scheme. Two such iterative schemes already exist: expectationmaximization, Bro et al. and parity check, Arciniegas-Alarc´on et al. The aim of this study is to present new variants of the basic method and to determine which iterative scheme produces the higher quality imputations. For this a simulation study was performed, and from incomplete matrices the quality of the imputations was assessed by estimating their uncertainty and by other criteria such as variance, bias and mean square error when a parameter of interest is considered. The best results were found using iterations with parity check and eliminating the singular values of the imputation equation.