The actual yields from multilocation variety trials are commonly analy
sed as a two-way table (rows = varieties, columns = locations) of mean
yields. In this paper, relationships between the recently proposed ne
w stability parameter 'variance of relative yield' and the two stabili
ty parameters of the 'classical joint regression approach in stability
analysis, are investigated. The variance of relative yield as a measu
re of stability exhibits an undesirable dependence on the mean yield o
f the respective variety. Generalizations and improvements of some rec
ent results and conclusions of Piepho (1994) are discussed.