Hi Frank, Your implementation says that you are implementing the third approach: % \item % Keep the definition expandable and fail if at the start of % \texttt{eqnarry} and the like. i.e. it will fail if the accented character occurs as the first character inside the cell: $$ \halign{$#$&$#$\cr ^^e4^^e4&b\cr c&d\cr } $$ here, the first ^^e4 will be "ignored" by TeX (with default fonts, the character will just silently disappear, with a message in LOG file: Missing character: There is no ^^e4 in font cmr10! well... my mathtext package tries to deal with this situation, and produce the characters even if they appear at the beginning of the cell... :) also, i'm not sure how would i define cyrillic letters via the \DeclareMathMeaning command, to preserve kerning... perhaps i have to use \DeclareMathSymbol for each cyrillic letter (having defined the math alphabet for cyrillic letters first), and then use that symbol inside \DeclareMathMeaning: - pretty much what is being done inside the mathtext package, but in a more compact way (perhaps one can redefine \DeclareTextSymbol too, to do all these things)... Please also look at my (very late) answer to your email on the mathtext package. For cyrillic texts, it is important to make the first character in a cell work (but not disappear), so we can try to use some extension to the approach used in mathtext package WRT \if@mmode hack - i mean that to make it 100% correct, some changes might be needed to the alignment preamble code... Of course, users can just use {} or \relax as the first element of a cell to solve all these problems with disappearing characters... but it would be nice (and possible) to make a "real" solution. i wonder, maybe eTeX has some replacement to \ifmmode test which will stop the look-ahead search for \omit but will not break kerning/ligatures... also, one noticeable difference between your approach and mathtext is that the latter defines math variants of characters depending on font encoding, while your variant defines only one math meaning... i need to think about this again (it was a long time ago)... Best, v.