This paper presents a methodological theory of public opinion research which we call the "steps-and-parts" model for polling. This theory is an analysis of polling, not an analysis of opinion. It deals with the researcher's role-actions in polling and not with the respondent's reactions in the interview. But any polling also involves the opinions or speech behavior of the respondents. The findings of any poll are always a complex product of the two factors: the unbiassed opinion of the respondents, and the polling instrument for observing it. The public's opinion as observed in a poll always has the variation of the demoscope, or instrument error, combined in unknown degree with the unbiassed variation of the opinions of the respondents. This model for polling seeks to standardize the instrument so that its error or biassing effects will be least and the variance of the public's opinion will be most of the total observed variance.