When dealing with comma separated data one of the modules Text::CSV and Text::xSV should be your first stop.

When dealing with uniqueness ('only the largest magnitude within each second'second') hashes should spring to mind. Combining those ideas and adding a little error checking (the various sample data you provided were inconsistent) the following should point you in the right direction:

use strict;
use warnings;
use Text::CSV;

my $csv = Text::CSV->new ();
my @expectedFields = qw(X Y Z Time Amplitude);

# Validate the header line
$csv->parse (scalar <DATA>);
my @fieldNames = $csv->fields ();

die "Unexpected field list: @fieldNames\nExpected: @expectedFields\n"
    unless @expectedFields == @fieldNames;

for my $fieldIndex (0 .. $#fieldNames) {
    next if $fieldNames[$fieldIndex] eq $expectedFields[$fieldIndex];

    die "Got field name $fieldNames[$_]. Expected $expectedFields[$fie
+ldIndex]\n";
}

# Find maximums in each 1 second slot
my %maximums; # Keyed by date/time

while (defined (my $line = <DATA>)) {
    $csv->parse ($line);

    my ($x, $y, $z, $time, $amplitude) = $csv->fields ();

    $time =~ s/\.\d{3}//; # Strip fractional seconds
    next if exists $maximums{$time} && $maximums{$time}{amp} >= $ampli
+tude;

    $maximums{$time}{amp} = $amplitude;
    $maximums{$time}{line} = $line;
}

# Output results ordered by time assuming the same date
print $maximums{$_}{line} for sort keys %maximums;

__DATA__
X,Y,Z,Time,Amplitude
2550,531,66,10-12-2007 07:03:08.069,2
2549,529,62,10-12-2007 07:03:08.151,1
2550,531,66,10-12-2007 07:03:09.069,1
2549,529,62,10-12-2007 07:03:09.151,2
[download]

Prints:

2550,531,66,10-12-2007 07:03:08.069,2
2549,529,62,10-12-2007 07:03:09.151,2
[download]

Note: always use strictures (use strict; use warnings;).