Tree analysis, widely used to evaluate the grade-recovery-boundary by generating a grade-recovery-curve, is used in evaluating froth flotation circuit separation efficiency. This paper analyses the tree analysis experimental procedure and then models it. It is suggested that since many output streams in a tree analysis are identical, a fact hitherto unreported, there may be large errors in the experimental procedures. Modeling suggests that there is no such thing as a unique grade-recovery-curve; rather, each such curve is dependent on the operator, the equipment used, how the experiment was run, and, more important, the level of the tree analysis, e.g., 4, 20 or 100. Clearly, the experimental tree analysis procedure is arbitrary: the surface chemistry varies during any one float and differs among floats in the same analysis, is sensitive to the decisions an operator makes, requires an experienced operator, depends on the discretion of the operator, has an arbitrary end-of-float point, is dependent on when and the amount of collector added, depends on the type of cell used, how that cell is operated and whether and how much froth cleaning spray is added. Regrettably, tree analysis is neither well defined nor readily repeated by independent operators. Two instrument designs for doing tree analysis are presented. The first automates tree analysis enabling 20, 50 or 100 level trees to be run. Unfortunately, this tree analysis instrument is difficult to stabilize. A second, more accurate instrument is also discussed, which is easier to operate.