We tested the reliability and validity of observer-rated pain in infants undergoing immunization using the visual analog scale (VAS). Pain was assessed in real time and later, from videotapes, in 120 1-year-old infants participating in a double-blind randomized controlled trial of amethocaine vs. placebo. Altogether, 2 (1 physician, 1 non-physician) of 4 raters [2 physicians, 2 non-physicians (nurse and graduate student)] independently assessed baseline and vaccine injection pain using a 100mm unmarked VAS line. Intra- and inter-rater reliability, assessed using the intra-class correlation coefficient (ICC), ranged from 0.69 to 0.91 and 0.55 to 0.97, respectively. Bland-Altman plots demonstrated no evidence of bias between real time and video assessments. When scores were dichotomized into 2 groups (no pain and pain) using a cut-off of >or=30mm, intra-rater reliability ranged from 0.35 to 0.92. The percent of scores deviating by >20mm was 4.5-14.29%. Criterion validity was demonstrated by correlations between the VAS and Modified Behavioural Pain Scale, a validated observational measure; (rho: 0.81-0.94). Injection scores were lower in the amethocaine group, when comparing difference (baseline-injection) or dichotomized scores; significance (p<0.036) was achieved for non-physician scores, but not physician scores. Together, these results provide initial support for the VAS as an outcome measure for acute procedural pain in infants. However, different conclusions may be reached about the effectiveness of analgesic interventions depending on the rater. Sources of variability include use of multiple raters, rater focus (procedure vs. child) and experience level.