Computer generation of random variables from a specified distribution is often done by first expressing the distribution as a mixture of a high-probability, easily generated element (or elements) and a lowprobability element.The high-probability element is generated by a direct (and fast) method; the low-probability element is often generated by the acceptance-rejection method.It is shown that for the usual applications of this well-known mixture-plus-acceptance-rejection (MAR) method there exists a corresponding application of the acceptance-complement (AC) method that is closely related to the MAR method.An analytic comparison shows that the AC method is slightly faster than MAR and just as easy to code.Illustration is provided by new algorithms for generating logistic variables.

Tile scales fl)r comput, er-printed graphs ~re determined by ~ simple algorithm providing a "reasonable" scale, i.e. one which centers the infc~rmation within the graphical frame and provides division points "along the scale which ~re simple numbers.The user m~y supply his own definitiml of simple numbers.Applications are to the construction of scatter diagrams, histograms and t, ables.A suggested formula for the number of interwfls is given which depends on the number of items to be displayed.

A pseudorandom normal number generator is proposed and its performance evaluated.Normal numbers are produced by first generating two uniform random numbers, each by a different mixed congruential generator, and then transforming them to two normal numbers.Eight million of these pseudorandom normal numbers are tested both for normality and randomness.The results of these tests indicate a close agreement with random normal theory.