window, "which was opened a certain amount during the tests"—a dangerous way, the author believes, to test radiators with the expectation of obtaining checking results—although "a screen was placed between the radiators and windows to prevent direct drafts from striking the radiators." These tests were unquestionably carefully made and were checked by reversing the position of the radiators, so that the comparative results obtained may be taken as reliable; but the coefficients were in the neighborhood of 20 per cent. higher than that obtained by the writer on similar radiators, due entirely to the greater freedom of air-circulation from the open windows. This is a matter of great importance in practice and in consequence the results obtained in radiator tests depend largely upon the setting of the radiator. It is this fact also which makes a radiator to some extent automatic or self-adjustable. Take for example, an ordinary room say with a north exposure and one direct radiator set, as they generally are and always should be, under the window. On a moderately cold day with the thermometer outside at 20 degrees and the wind from the south or east, the radiator will be turned on the entire time. On another day with the thermometer outside at zero and the wind from the north, the radiator very probably keeps the room at the same mean temperature with practically the same temperature of steam. The reason is that in the latter case the cold air which leaks in through the walls and around the window casing, besides keeping the air immediately in contact with the radiator at a lower mean temperature, causes a much more rapid circulation around the radiator, with the result that it gives out more heat units per square foot. It is for this reason, too, that one may put down for a positive and infallible rule in radiator design that the radiator which has the most open space around its surfaces and the most interrupted exposure to the surrounding air will give out the most heat per square foot under the same conditions of setting. In compliance with this rule, other things being equal, narrow radiators are more effective than wide ones, and low ones than high ones; but the effect of width and height can more than be offset by a slight increase in the distance between the loops; for example, the author found that a four-column 38-inch radiator gave out exactly the same heat per square foot of measured surface as a two-column 38-inch radiator, because the former had a mean distance between the loops about 16/100 of an inch greater than the latter; also a
