ATOMIC WEIGHTS. 533 



subject and worked out in a very similar way, although the 

 experiments are not so numerous. The experiments are 

 arranged in two series, in the second of which a notable 

 improvement in the arrangement of the apparatus is made, 

 viz., the introduction of a tube containing metallic copper 

 kept red hot so as to remove every trace of oxygen from 

 the hydrogen, and this is, of course, followed by drying tubes. 

 Four experiments were made in each series, in the first of 

 which the weighings were all reduced to vacuum standard, 

 but in the second series they were actual vacuum weighings. 

 In F. W. Clarke's recalculation of atomic weights (7) the 

 atomic weight of oxygen is given as 15*9369 + '0138 from 

 the first series and 16*0095 ± '0030 from the second. 

 Clarke's remarks on these results are not quite fair, especi- 

 ally as the experimenters themselves say that they regard 

 the second series as more accurate than the first. He says : 

 "The effect of discussing these two series separately is 

 somewhat startling. It gives to the four experiments in 

 Erdmann and Marchand's second group a weight vastly 

 greater than their other four and Dumas' nineteen taken 

 together. For so great a superiority as this there is no 

 adequate reason ; and it is highly probable that it is due 

 almost entirely to fortunate coincidences rather than to 

 greater accuracy of work.'' If, after making most important 

 improvements in their apparatus so as to obtain purer 

 hydrogen (especially as regards oxygen), as well as modify- 

 ing their manipulation so as to obtain the true weights of 

 oxygen and of water directly, to say nothing of their 

 experience, better and more concordant results had 

 not been obtained, in what direction could we look for 

 such ? 



Two other researches carried out in almost exactly the 

 same way as the above are of the highest importance and 

 interest, especially that by Dittmar and Henderson (8), 

 who intended at first only to try to detect the sources of 

 error to which the method was liable, but in the end made 

 a series of experiments w r hich seem to be free from the 

 many sources of error which they proved to exist in the 

 experiments of Dumas. The sources of error which had 



