Simulated data derived from random numbers are used to show that the process of relating u& and similar properties to the stability parameter z/L is highly susceptible to error. An alternative method, making use of Ri as a stability index, is not affected in this way and is used to re-examine the data obtained in the 1968 Kansas micrometeorological experiment. The relationship u,/u. N 1.25 (1 -2z/L)'13 is found to provide a good fit to the unstable data, but it is unclear as to whether a small peak observed in stable conditions is real (perhaps associated with gravity waves) or not (possibly a consequence of measurement errors). The properties 0,/u. and u&. are found to attain a relatively constant value (1: 3) in conditions more unstable than about Ri = -0.4. The 'shape' ratio u,/u, is found to decrease to less than unity in very unstable conditions, possibly as a consequence of some undetected error in measurement of uu. In the case of temperature fluctuations, u,/T. = 0.95 (-z/L)-"~ is found to provide an excellent lit in unstable conditions (Ri < -O.l), even though this form also agrees well with random behavior.