The present study examined consistency of temperament across situations and time in early childhood. On two occasions, 7 months apart, we observed the responses of 94 four-year-olds to multiple standard laboratory procedures eliciting exuberance, fear, anger, sadness, or interest. In addition, fathers' and mothers' reports on children's temperament were obtained. Observed temperament showed substantial context specificity, varying across situations within dimensions. Structural equation models revealed separate observation and parental perception factors and factorial invariance and high stability for most dimensions. Stability coefficients based on correlations varied according to level of aggregation. Measurement convergence was generally moderate and was lowest for negative emotionality. Implications for the assessment of temperament are discussed in relation to differences in consistency across dimensions and convergence between measures.