Two trials, conducted in two locations in Italy over 4 years with Chardonnay grapevines trained to GDC, Pergola or Casarsa systems and grafted onto 6 rootstocks (SO4, Kober 5BB, Teleki 5C, 8B, 140Ru and 1103P), provided vine yield and pruning weight data to study a practical and reliable method for assessing genotypic stability. Commonly used stability statistics, like indices based on coefficients of variation and regression, were estimated. When the deviations from regression of genotype performances on the environmental index were large, biological response of rootstocks to different environments was better fitted by double regression analysis based on Verma's coefficients. The unpredictable part of the response in regression approach was quantified by Wricke's ecovalence index. The results confirmed that the use of more than one statistical procedure is necessary in the selection of stable genotypes. Genotype × environment interactions were partitioned into four components (variability, ecovalence, linear response or deviation from linear response, environmental sensitivity). The use of a graphical representation of the four components was easy to use and appropriate for grouping genotypes of greater and lesser stability. Rootstocks Teleki 5C and 8B were identified as stable for grape yield, while high yielding genotypes (SO4 and 1103 P) were judged as unstable. 140Ru and Kober 5BB were detected as stable for yield of pruning wood.