Stability statistics that are used to explain genotypic response to en
vironments are not useful to plant breeders unless they are repeatable
across sets of environments. The purpose of this study was to examine
the repeatability of the stability estimators: regression coefficient
(b(i)), regression coefficient away from mean regression (\b(i) - 1 \
), mean squares for deviation from regression (Sd(i)2), Shukla's stabi
lity variance (sigma(i)2), variance of genotypic means (S(i)2), and ge
notypic coefficient of variation (CV(i)), in addition to coefficient o
f determination (r(i)2) and mean yield (X(i)BAR). These statistics wer
e calculated from three sets of yield data of the Louisiana Agricultur
al Experiment Station winter wheat (Triticum aestivum L.) performance
trials grown in 36 environments. Repeatability was estimated by Spearm
an's rank-correlation coefficient and Kendall's coefficient of concord
ance. The \b(i) - 1\, sigma(i)2, and Sd(i)2 were not repeatable betwee
n any two subsets of environments, and repeatability of S(i)2 and r(i)
2 s were low. Among the stability estimators, only b(i) and CV(i) were
repeatable across subsets of environments. The CV(i) was not a reliab
le statistic to describe genotypic stability because the rank order of
CV(i) was induced by the rank order of X(i)BAR. Mean yield was the mo
st repeatable genotypic character. Gain in selection for yield stabili
ty can be expected from the combined use of b(i) and X(i)BAR. If data
do not fit the linear regression model, then low values of S(i)2 (Type
1) or Type-4-variance (variance of genotypic means across unpredictab
le environments averaged across predictable environments) may be used
as an alternative criterion for yield stability.