We have discovered a severe problem with the current theory of species
-area curves (SPARs). This theory claims that we should expect SPARs w
ith z-values of about 0.26. However, that is wrong. The correct predic
tion turns out to be approximately 0.77. To make this prediction, we u
sed a stochastic sampling scheme, and constructed species-area curves
from a lognormal abundance distribution, exactly as previous theory me
ant to do. We arrived at our prediction using two independent methods:
we performed computer simulations of the scheme and we derived its an
alytical equation. SPARs that result from the simulations are the same
as those from the equation, validating the logic of our analysis. We
explain what went awry with the previous theory. However, although log
ically accurate, the new theory has an empirical problem: real SPARs d
o not have z-values near 0.77. Rather, they tend to lie in the interva
l 0.1-0.2. To obtain these, we added an assumption to the lognormal ab
undance distribution. We assumed that range size and abundance are pos
itively correlated. This new assumption is qualitatively similar to Ha
nski's (1982; Oikos 38: 210-221) pattern. Finally, we derive a simple
relation connecting average point diversity, average range size and sp
ecies diversity for a province.