Interviewer effects in recording can bias survey data in two ways: (i) unsystematically, by increasing random error; and (2) systematically, by suppressing natural respondent variability. Many studies have dealt with random error, but few with systematic error. This study investigates the problem of systematic bias from recording in survey data. Unsystematic error is typically related to incompetence, or the interviewer's lack of the clerical and critical skills needed for recording. Guest and Nuckols, Sheatsley and others have identified certain personality and demographic correlates of competent recording (i.e., recording with relatively few clerical errors); e.g., intelligence, sex, marital status, education, and experience in interviewing appear to be good predictors.' These unsystematic effects are generally considered endemic to survey research. Random error is a calculated risk, taken as the better part of the implicit trade-off with unsatisfactory features of other datacollection methods. Generally, the hope is that instances of random error will balance out over many interviews. Systematic effects cannot be so hopefully dismissed. Typically, they result from consistent interviewer behaviors, or idiosyncrasies. These create less variability within one interviewer's production than between sets of schedules submitted by several different interviewers. By contrast, in the ideal situation where each interviewer's quota of respondents represents a random subsample, the between and within variances should be approximately the same. For example, an individual interviewer can be expected to use words in a systematic way. If she does not record her respondents' words verbatim, then neces-