We could have had a computer center perform these analyses; why do it ourselves? (1) Time is one of the considerations. The tum-around time not only includes delivering the data to the computer center, but also communicating what is to be done. (2) We really wanted to operate on these data in an interactive way, i.e., depending on the outcome of an analysis on a subset of variables, we would try various subsets. This is not interactive at the level of the experiment, but rather at the level of data analysis. (3) Computer centers prefer to have data come in on cards, and ours were not on cards. Therefore, considering the long chain of people and expense of the machine involved , it seems less prone to error and cheaper to do whatever analyses we can ourselves. large group of Ss. Using FORTRAN, we wrote a program which gave us the correlations among all the variables . In examining the correlation matrix, we saw some interesting patterns which led us to want a multiple-regression analysis that could combine the questionnaire variables in an equation best predicting the EEG scores. We had not written a multiple-regression program and had only a few days left before a meeting. We dipped into the IBM stockpile for the appropriate subroutines, which we linked together with a FORTRAN program. The program read our correlations off our disk (where they had been stored by our correlation program) and typed the results of the multiple regression on the Teletype relatively quickly . • eletype) cost less (approximately $45 ,000) than a small-scale system bought about 10 years ago (4K 22·bit delay-line memory, A/D with 7-channel multiplexer, D/A with 2 channels, 22 digital input-outputs, paper-tape reader, and fIexowriter). Furthermore, the cycle time is 10 to 1,000 times faster on the 18/30. Most of what we have to say applies to a variety of computers. However, one aspect of the GA 18/30 is relatively unique : its command structure has been patterned after two IBM machines from which the 18/30 gets its hybrid name. The IBM machines are the 1800, a process controller discussed in the next talk, and the 1130, which was designed as a batch processor. The GA 18/30 has the same commands plus another class of register-to-register commands. This means that the library of routines developed for those IBM machines may be used on the faster GA 18/30. As one example, we had obtained a large set of data (EEG scores from a special -purpose EEG scorer and answers to a questionnaire) from a We are attempting to handle two classes of problems with our General Automation 18/30 system (Anaheim, Calif.): (l)to use the computer directly at the level of the experiments (a) to control the experiments and (b) to collect the data; and (2) to use the computer with the data from experiments, i.e., to analyze the data and make theoretical calculations. These two classes of problems are similar to those which, in other contexts , have sometimes been characterized as (1) process control and (2) batch processing. At the level of the experiments, for example, we want the system to t ime and present various visual stimuli according to a particular experimental paradigm while the S's physiological responses (EEG, averaged evoked potentials, EKG, etc.), behavioral responses (switch closures), and control pulses are being collected. Some of our research led us to elaborate statistical analyses, and these large problems forced us to go beyond the capacity of the small computer system we have been using (a Packard Bell 250). Our difficulties in dealing with large computer centers reinforced the notion of attempting to perform as much computation on site as possible . This was one of the considerations in setting up a medium-size computer system rather than staying with a small system. Another consideration was the lowered costs of computers, especially the central processors. It is possible to obtain a four th-generation medium-size central processor at about the same or lower cost than previously designed small-scale computers and get more powerful organization and command struc ture . For example , our GA 18/30 system (8K 18-bit core , 512K-word disk, 8 priority interrupt levels, 5 direct memory-access channels, 3 programmable timers, A/D with 16-channel multiplexer, D/A with 8·channel output holding amplifiers, 16 digital inputs, 16 digital outputs, paper-tape reader, card reader, and