Introduces a linear regression method for investigating unconscious cognition. For words that were obscured by simultaneous dichoptic masking, indirect effects (semantic priming) and direct effects (perceptual identification) were assessed in 20 experiments (total N = 2,026). When measures of both indirect and direct effects have rational zero points, a statistically significant intercept in the indirect-on-direct-measure regression shows that (a) the indirect effect occurred in the absence of the direct effect, and (b) unconscious cognition is involved. For a position discrimination task, but not for an evaluative decision task, indirect-on-direct regression showed the significant intercept effect. Although small in magnitude, this intercept effect provides the statistically most secure finding yet obtained of a much-sought and controversial data pattern--indirect effect with no direct effect. With one added assumption (which appears plausible for the present data), this pattern indicates that unconscious cognition is dissociated from (i.e., occurs separately from) conscious cognition.