According to major cognitive theories of emotional disorders cognitive biases are partly responsible for their onset and maintenance. The direct test of this assumption is possible only if experimental method capable of altering a given form of cognitive bias is available. The purpose of the study was to examine the effectiveness of a novel implicit memory bias training procedure based on the emotional version of the classical Sternberg’s short-term memory task with negative, neutral and positive words. 108 participants, who completed the PANAS and the CES-D questionnaires, were randomly assigned to the control group (n = 33), the No-Negative group (n = 36), in which the target words in the Sternberg’s task were either positive or neutral but never negative or the Negative-New group (n = 39) in which the negative target words in the modified Sternberg’s task were always new. This training was followed by the recollection stage. Only one of the training protocols resulted in significant effects at the recall stage - individuals in the No-Negative group recalled more positive words and fewer negative words than those in the control group. These results show that it may be possible to experimentally induce memory bias characteristic of non-depressed individuals.