The latest Earnings Mean estimate right before the Fiscal Period Ending date is the one you want.
In summary statistics page, if you query for IBM, Forecast Period = "Fiscal Year 1", you'll get the following results:
OFTIC MEASURE FPEDATS FPI STATPERS FISCALP NUMEST MEANEST
IBM EPS 31DEC2002 1 17JAN2002 ANN 18 4.81
IBM EPS 31DEC2002 1 14FEB2002 ANN 21 4.78
IBM EPS 31DEC2002 1 14MAR2002 ANN 22 4.77
IBM EPS 31DEC2002 1 18APR2002 ANN 20 4.11
IBM EPS 31DEC2002 1 16MAY2002 ANN 21 4.10
IBM EPS 31DEC2002 1 20JUN2002 ANN 21 4.06
IBM EPS 31DEC2002 1 18JUL2002 ANN 18 3.99
IBM EPS 31DEC2002 1 15AUG2002 ANN 20 4.00
IBM EPS 31DEC2002 1 19SEP2002 ANN 20 3.96
IBM EPS 31DEC2002 1 17OCT2002 ANN 21 3.92
IBM EPS 31DEC2002 1 14NOV2002 ANN 20 3.91
IBM EPS 31DEC2002 1 19DEC2002 ANN 22 3.91
The last mean estimate dated 19Dec2002 is the one you want.
Moreover, if a company reports earnings near the end of the month, the announcement will probably come after an IBES Statistical Period (usually in the middle of the month). In that case, there will be a consensus estimate *after* the Forecast Period End Date but *before* the Report Date of earnings that will be the last valid consensus estimate before earnings are reported. For examples of this, use the date range 2001 - 2003 for the Tickers "F", "T", "XOM".
If you use the "Summary + Actuals" web query, you can just pick out the last record for a given FPEDATS. For a given Forecast Period End Date choose the record with the latest Statistical Period date that precedes the Announce Date. That's the prevailing consensus estimate before earnings are reported, which may come after the Forecast Period End Date.