Warmth and competence are two basic dimensions in social judgment. Warmth having traits related to perceived intention such as friendliness, helpfulness, sincerity, trustworthiness and morality, while competence having traits related to the ability to achieve the intention, such as intelligence, skill, creativity and efficacy. Recent researches suggested that warmth dimension could be divided into sociability and morality. Previous researches showed much evidences for both positive and negative inferences between warmth and competence. Previous researches also indicated that context cues played important role in the social inferences between warmth and competence. The present study tested how context cues influenced the references between different social cognition contents. Study 1 tested the positive inferences between warmth and competence under conditions regardless of context. In study 1, 38 university students participants were assigned to a 2 (description：warmth vs. competence) × 2 (valence: positive vs. negative) within-participant design. Participants needed to rate four types of targets, whom were descripted as either positive or negative on one dimension of warmth and competence. The results of study 1 showed that the positive descriptions on one dimension would trigger positive inferences on the other dimension, which suggested the positive inferences between warmth and competence under the conditions without context cue. Study 2 furthermore explored the references between warmth and competence under two different context conditions. 40 participants were recruited on campus and assigned to a 2 (context cue: point to high competence vs. do not point to high competence) × 2 (description：high warmth vs. high competence) within-participant design. The results showed that, when the context cue pointed to the high competence clearly, which suggested that this situation need a target with high competence, the positive description on warmth elicited a negative inference on competence and negative behavioral reactions toward the target, whereas the positive description on competence elicited a correspondently positive inference on warmth dimension and positive behavioral reactions toward the target. However, when the context cue didn’t point to the high competence, both warmth and competence descriptions could elicit positive inferences on the other dimension. In study 3, 74 female participants were randomly assigned to a 2 (target occupation: nurse vs. judge) × 2 (description：high sociability vs. high morality) between-participant design. The results showed that the targets would get better evaluations if descript the sociability of nurse or descript the morality of judge. Together, the present study found that: 1) When lack of context cues or the context cues were ambiguous, participants would make positive halo inferences between warmth and competence dimensions; 2) When the context cues point to clear cognition contents, the positive descriptions on the corresponding cognition content would elicit positive social inference. The present study suggested that context cues played important role in social inference and the targets would get better evaluations if the descriptive information matched the context expectations.