Introduction Since Stigler's (1962) seminal article on information in the labor market, economists have studied screening. Most of this literature has examined screening in the context of a firm's hiring decision. That is, firms cannot perfectly observe the productivity of their workers prior to hiring them, but they can attempt to screen out less productive workers if they know that some observed attribute (e.g., schooling) is highly correlated with productivity. There has been a great deal of empirical work on educational screening (see Riley, 1979, for a review). More recently, the relationship between firm size and screening has come under closer scrutiny (see, e.g., Garen, 1985; Barron, Bishop, and Dunkelberg, 1985; and Barron, Black, and Loewenstein, 1987). There are, however, other contexts in which firms might engage in screening, an important one being an insurance market.(1) In an insurance market, firms must confront individuals who are good risks and individuals who are bad risks, but they cannot directly observe the riskiness of a given individual. Rather, they can attempt to use information on the observed attributes of individuals to determine riskiness and on that basis insure only the individuals perceived to be good risks, leaving the bad risks to be insured by other firms or by an assigned risk plan.(2) If only certain firms successfully screen out bad risks, these firms would have lower loss costs and lower premiums for a given insurance policy than other firms. Thus, price dispersion in automobile insurance, which Dahlby and West (1986) found to be consistent with a model of costly consumer search, can also be consistent with differential screening. This article follows Dahlby's (1988) modification of the Carlson and McAfee (1983) model of price dispersion to allow for screening. Unlike Dahlby, however (but consistent with Barron, Bishop, and Dunkelberg, 1985), effective screeners are predicted to consist primarily of large firms, and they should insure fewer high-risk drivers and have lower loss costs as a result. We examine the relationship between firm size and loss costs using automobile insurance industry data for Alberta for the years 1978 through 1981. The empirical results generally support the implications of the theory. In particular, the relationship between cars insured per firm and loss costs per car insured is significant and is represented by a parabola, as expected, for four out of the five largest driver classes in Alberta. The next section briefly reviews the Carlson and McAfee model and presents an extension that incorporates screening. The testable implications relating to screening are also derived. In the subsequent section, we describe the data that are used in the tests for screening, and some descriptive statistics are presented. Then we discuss the empirical results and, finally, provide a summary and some concluding remarks. Screening in a Model of Price Dispersion This study's tests for Screening in an automobile insurance market are based on an extended version of the Carlson and McAfee (1983) model of price dispersion.(3) Assume that there are J driver classes, j = 1, 2 ,..., J, defined on the basis of some easily observed characteristic of a driver, such as age or sex. (Since the following analysis applies to each driver class, the j subscript will be suppressed.) There are M consumers, and each consumer has one car that must be insured. There are N firms, indexed i = 1, 2 ,..., N, and firm i's premium is [z.sub.i], with [z.sub.1] [less than or equal to] [z.sub.2] [less than or equal to] ... [less than or equal to] [Z.sub.n]. The probability that a consumer will observe firm i's premium after engaging in one unit of search is 1/N. This implies sampling with replacement. Assuming that consumers know the distribution of premiums and use a sequential reservation price search strategy and that the distribution of search costs across consumers is uniform over the interval [0,2Y], Carlson and McAfee have shown that the demand curve facing firm i is [Mathematical Expression Omitted], where [Mathematical Expression Omitted] = the quantity demanded from firm i, [R. …