A two-choice, forced-choice method is used in which a single trial consists ever attained the higher level of per- of the presentation of two light-sound pairs. For the standard pair, a circular spot of light and a 2000-hertz pure tone begin simultaneously and terminate simultaneously. For the variable pair, quantum state, And in every case a one-quantum function has been obtained in later sessions. A subject who enters a two-quantum State typically remains in that state for many sessions. the signals begin simultaneously but the light is terminated ¢ msec before the sound, The rationale for using simultaneous terminations for the standard is discussed elsewhere (5). On half the trials, chosen at random, the standard Switching to is presented first; on the remaining trials the variable occurs first. The subject responds by indicating whether the offset of the light occurred before the offset of the tone in the first pair or in the second pair. When he indicates the variable he is scored correct and the proportion of correct responses is de- termined for each of a number of values of ¢. This proportion, P(c), as a function of ¢, is the successiveness discrimination function. The shorter member of each pair of signals has a duration of 2.0 seconds and the empty interval between pairs is also 2.0 seconds. An auditory signal immediately following the subject’s responsetells him whether the response was correct, Trials are run at the rate of one every 12 seconds. Usually only one 20-minute session is conducted each day for each subject. Equal numbers of the different values of ¢ are randomly assignedtotrials. Under some conditions the successiveness discrimination function can be described adequately by a single straight-line segment (6). A line fitted to data intersects the level of chance performance; that is, P(c) = 0.5, at a value of t which is called x. The line reaches P(c) = 1.0 at a value of ¢ of (x + M) msec. The parameter M is the minimum amount of time which must be added to the interval between the light and sound offsets to bring P(c) from 0.5 to 1.0. I have reported (3) an average value of 54 msec for M from data obtained by using highly practiced subjects under conditions which were designed to maximize performance. This value was suggested as one estimate of the duration of a quantum. However, with several subjects I have observed a strikingly different result under certain conditions. Instead of a “one-quantum”function they yield data which, while still linear in form, span about 100 msec. These “two-quantum” functions have always been observed early in practice, before the subject has 1338 formance associated the with the one- one-quantum state ordinarily requires some change in the experimental procedure, These observations suggested the pos- sibility that there are two distinct states such that when a subject is in state 1 his value of M is one quantum and when he is in state 2 his M is two quanta. Further, under some experimental conditions the probability of being in state 2, P2, may be close to unity, while under other conditions the probability of being in state 1 may be close to unity. If this conception is correct, then it becomes apparent that the earlier measurements of M were based on the limited to the first quantum above x and most subjects did not quite attain a P(c) = 1.0 even for the greatest value of ¢, Therefore, a new experiment was done to test the adequacy of the two-state model. The ten values of ¢ from 30 to 120 msec in 10-msec steps were used and the interval between offsets for the standard was fixed at 20 msec. All of these intervals have a positive sign which indicates that the light preceded the sound. Twenty-three young, adult subjects participated, 14 male and 9 female. We eliminated four of them at the beginning because we were unable to cbtain alpha in their electroencephalograms. The remaining 19 were run through One session per day and an electroen- cephalogram was taken before and after each session. Analysis of the electroencephalogram consisted of selecting mean of two linear functions having the monorhythmic single cycles of alpha and measuring the period of each to the nearest millisecond under a comparator. Twenty such samples were measured in each record, giving forty measurements per session. All of this analysis was performed by assistants who had no knowledge of the psychophysical analysis (7). Obtained values of P(c)} are presented in Table 1 for 13 subjects. The other six subjects are not analyzed one case and two in the other. The model failed to fit and five because they assumption that a state 1 probability of unity was actually achieved. To the extent that this condition was not met, values of M7 would exceed the duration of one quantum. These considerations lead to a two- state model of successiveness discrimi- nation in which the successiveness discrimination function is the weighted same x but spanning one quantum in weighting factor is P2, the probability of being in state 2. The function consists of the following four regions: when: (x—M) <t<x P{c) equals: 0.50 xetc@tM) “GPa-osp,) + 0.5 oe W+M)=t (tf — x)P2 ai" + < (x -+2M’) (1 —0.5 P.) t= (x + 2M’) 1.00 in which M’ is the quantum size in milliseconds. This two-state function rises from P(c) = 0.5 to P(c) = 1.0 as two linear segments, one connecting the points (x, 0.5) and (x + M’, P) and the other connecting (x + M’, P) and (x + 2M’, 1.0). The intersection of these segments occurs at P(c}) = P= 1 — 0.25 Po. To apply this model requires obtaining data points over the range of t between x and (x + 2M’). In the earlier experiments this range was further, one because did not reach P(c) the two-state > 0.90 even at t = 120. An even wider rangeof values of t should be used. The combination of the three parameters of the two-state model which yields the minimum squared-error fit was determined for each of the subjects. For this solution, values of x and M’ were found to the nearest millisecond and P was determined to the nearest one hundredth, The results of these computations are listed in Table 2 along with the modal (peak) alpha interval. , The two-state function fits the data satisfactorily, as Fig. 1 demonstrates. This figure is a composite of all the subjects with each one entered in rela- tion to his own parameters as explained in the caption of Fig. 1. The two segments of the function are both described adequately by the model, consistent with the deductions that there are two segments and that they span equal distances on the abscissa, The quantities M4’ and alpha are very similar. They have the same mean, although the standard deviations suggest that M’ is somewhat more variable. SCIENCE, VOL. 158