The initial reports in this issue of two new studies of childhood brain cancer and 60-Hz magnetic fields (1,2) illustrate a few general observations on evolving research in this area. Epidemiologic assessment of magnetic field exposures is still primitive. Magnetic field exposures come from many different sources. We should not be proud that we still study them one by one. Appliance-specific relative risks (one for blow dryers, one for curling irons, etc.) are the most glaring example. Time-activity data and magnetic flux density measurements are needed to quantify exposures from individual appliances. At a minimum, this information could be used to stabilize ensembles of appliance-specific relative risk estimates (3-6). Ideally, appliance exposures would be consolidated and combined with measures of exposure from power lines and other sources. Developing a program of combined-source magnetic field exposure assessment tailored for use in epidemiologic studies would be a challenging, but technically feasible, task. No methodological need is greater in this area of research. Responding to a private exhortation to work toward a multiappliance exposure measure, an epidemiologist once expressed to me a reluctance to apples and oranges. The simile was ironic. Combining fruits is exactly what is done in the more mature field of nutritional epidemiology to measure consumption of fructose and other nutrients. For power-line exposures, it has become clear that codes, i.e., magnetic field exposure measures based on characteristics of lines and their distances from residences, have substantial advantages over short-term flux density measurements in homes. The measurements, often made years after the case or control actually lived in the residence, are obviously limited by their temporal variability. Independently severe is the problem of missing data. Whether by design (1) or because of nonresponse (7), it is not uncommon for magnetic field measurements to be obtained for fewer than 50 percent of cases and controls. Relative risk estimates based on such small numbers of subjects are of highly limited value, whether viewed in isolation or in contrast to estimates based on wire codes. The in-home measurements should not be abandoned entirely, however. They have considerable value when used in conjunction with wire codes rather than as alternatives to them. Specifically, the measurements can provide exposure scores for wire code categories in dose-response analyses (8). Conducting such an analysis forces one to confront the fact that not all wire code categories appear to differ in exposure. If any principle of dose-response analysis is beyond dispute, it is that response should not vary when dose does not vary. It is therefore necessary to combine some of the wire code categories. The categories to combine may change from locale to locale and over time within a locale. In the report of their Los Angeles County study, Preston-Martin et al. (1) laudably focus on this issue. The authors present the case and control counts in each of the five wire code categories, but combine the second and third categories (very low and ordinary low) for data analysis. Only an anomalistically elevated relative risk estimate in the lowest (underground) category prevented them from combining the first three categories. Alternative analytic approaches could be defended, however. In the extreme, some of the more exposure-relevant measurements, e.g., median 24-hour child's bedroom values of 0.5, 0.6, 0.4, 0.6, and 1.1 mG, might have supported a decision to combine all categories but the highest. Unfortunately, the paper by Gurney et al. (2) does not provide quantitative information on magnetic flux density levels within wire code categories in western Washington State. Without such information, one cannot tell how strongly to interpret the resolutely null results the authors report for childhood brain cancer across the five wire code categories in that study.