What is your formula in L5 and what values are used by that formula, not
formatted, but actual?

Tyro

Unfortunately, this inconsistency is intended by MS, and will occur anytime
that L5 is not an integer but differs from INT(L5) beyond the 15th
significant figure (such as =1+2^-52).  Under those circumstances expressions
like =L5-INT(L5) and =L5=INT(L5) cannot be relied upon to accurately reflect
the value in L5.  Instead you need to structure the formula such that the
subtraction or comparison is not the last operation, as in =(L5-INT(L5)) or
=L5-INT(L5)<>0.

As documented, Excel will display no more than 15 significant figures.  If
you ask for more, you will get a result padded with meaningless zeros instead
of an accurate reflection of the cell contents, but you can indirectly
determine what is going on
 http://groups.google.com/group/microsoft.public.excel/msg/6efb95785d1eaff5

Almost all software does math in finite precision (the only alternative
would be symbolic math, which would be prohibitively slow in a large
spreadsheet).  Under some circumstances, finite precision math necessarily
violates familiar mathematical rules such as x = a+x-a.  For example, if you
were doing 4-figure decimal math, then
 1+1/3-1 = 1+0.3333-1 = 1.333-1 = 0.333 <> 0.3333
Most people recognize that 1/3 has no finite decimal representation, and so
are not surprised by this, but people who have not thought about the binary
math underlying Excel and almost all other software are surprised when
similar things happen involving numbers like 0.1 (which has no finite binary
representation).  In an effort to smooth this out (a technically impossible
challenge) MS introduced an "optimization" in Excel 97
 http://support.microsoft.com/kb/78113
such that final subtractions of numbers that are equal to 15 decimal digits
will return zero, and final comparisons of numbers that are equal to 15
decimal digits will be declared equal.  This will only be an improvement if
the small real differences only exist because of finite precision
approximations.  The "optimization" cannot be extended beyond the final
operation, because it would result in increasing inaccuracy.  Depending on
what calculations have taken place, this fuzz factor may be too much or may
not be enough--one size does not fit all.  A better approach would have been
to educate users with guidelines on how to more effectively get the intended
results in finite precision, but there have been 4 subsequent Excel versions
with no indication that MS recognizes the need to rethink their approach, so
I won't hold my breath.

The classical discussion of the issues around finite precision floating
point arithmetic is Goldberg's article "What every computer scientist should
know about floating point arithmetic"; copies can be found many places on the
internet, such as
 http://www.physics.ohio-state.edu/~dws/grouplinks/floating_point_math.pdf

Jerry