1. Aggression experiments were performed with Gnathonemus petersii. Three kinds of stimulus fish were used in random order to elicit attacks: (1) intact conspecifics, (2) electrically silent (operated) conspecifics, and (3) the mormyrid Brienomyrus niger.
2. The aggression indices for the two experiments with B. niger and the intact conspecific, i,e. the two electrically active fish, were similar (p > 0.4), whereas the index for the experiment with the silent conspecific was about one half of the two others (p<0.01, Fig. 2).
3. In none of the three experimental conditions can the behavior sequence displayed by G. petersii be described by a random model, i.e. the occurrence of a given act depended on at least the preceding act (p<0.001, Tables 1-3), as in a first-order Markov chain. For every behavior recognized, the
distribution of following acts on a given act found in one group of experiments was significantly different from the same distribution found in one of the two other groups of experiments (Table 4).
4. The behavioral sequences, displayed by the experimental subjects, exhibited redundancies which were only slightly different when an intact (35%) or an electrically silent conspecific (33%) were present (Table 5). A considerable
difference between both experiments was found on comparing the information T(x; y) which is common to a preceding behavior x and a following behavor y: in the silent conspecific experiment T(x" y) was more than twice the value found in the intact conspecific experiment (0.65 and 0.28 bit, respectively). In the B. niger experiment T(x; y) was found to be low (0.24 bit). In contrast to both conspecific series of experiments, the redundancy value was also low (23%). A high degree of redundancy of the behavioral sequence in spite of a low amount of internal constraint-both considered to be of prime importance in a communication system-were shown by the experimental G. petersii in the presence of intact conspecifics only.
5. There were considerable differences in the electrical discharge activities of B. niger and G. petersii (Figs. 4-6). B. niger did not display a Preferred Latency Response to G. petersii's pulses (Fig. 7).
6. The results show that by conspecific electric signals information is transmitted which significantly influences the behavior of a recipient G. petersii. These signals seem 'to be strictly species-specific.