The correlation of the baseband frequency response of single lengths of multimode fibre with that of an assembled set of such fibres has remained one of their most unsatisfactory features. Many approaches to this problem have yielded only partial success in a predictive sense, despite their effectiveness in accounting, after the event, for observed experimental results. The common feature among most predictive procedures is an assumption of 'good behaviour' of the fibres involved. Others have attempted to take into account 'rogue behaviour' (fibres having modes separable into distinctly fast and slow groups) by manipulation of the transfer function, in either time or frequency domains, in an effort to derive a frequency parameter leading to a consistently applicable predictive theory. The motivation, apart from scientific thirst for knowledge and understanding, has been economic. An algorithm enabling a required system performance to be achieved with certainty, without an excessive margin, and using the most readily available, lowest-cost fibre would clearly be a desirable outcome. Has such an outcome been attained? If so, to what extent, and under what conditions? Does the complexity of the available algorithms in any way detract from their practical utility and economic benefit? Are there better, simpler ways of achieving the end result? The author is of the opinion that the problems, scientifically fascinating though they are, are likely to remain largely unsolved, because in the end, they will not be worth solving.