As argued in Part I (Derbyshire, 1995), variability is a key issue in
stable boundary layers, and differences in variability between observa
tions and idealized models may imply sytematic biases. Here we discuss
how data analysis can be geared to allow for variability and thus con
sistency with models. Instrumental errors, smoothing methods and verti
cal discretization are considered. We then show how statistical averag
ing broadly improves the agreement of 'heterogeneous' results in Part
I with the Brost-Wyngaard closure. Recommendations are made for the in
formation needed to analyze apparent differences between 'homogeneous'
and 'heterogeneous' stable boundary layers.