Coercing the value to float and back to string was a neat trick. You can also just add a literal 0 to whatever you're checking.

<?php
function isfloat($value) {
  // PHP automagically tries to coerce $value to a number
  return is_float($value + 0);
}
?>

Seems to work ok:

<?php
isfloat("5.0" + 0);  // true
isfloat("5.0");  // false
isfloat(5 + 0);  // false
isfloat(5.0 + 0);  // false
isfloat('a' + 0);  // false
?>

YMMV

If you want to test whether a string is containing a float, rather than if a variable is a float, you can use this simple little function:

function isfloat($f) return ($f == (string)(float)$f);

is_float() returns true for NAN, INF and -INF. You may want to test is_float($value) && is_finite($value), or alternatively filter_var($value, FILTER_VALIDATE_FLOAT) !== false.

As celelibi at gmail dot com stated, is_float checks ONLY the type of the variable not the data it holds!

If you want to check if string represent a floating point value use the following regular expression and not is_float(),
or poorly written custom functions.

/^[+-]?(([0-9]+)|([0-9]*\.[0-9]+|[0-9]+\.[0-9]*)|
(([0-9]+|([0-9]*\.[0-9]+|[0-9]+\.[0-9]*))[eE][+-]?[0-9]+))$/

Boylett's solution is elegant (http://php.net/manual/en/function.is-float.php#85848), but won't work for long float's or variables that are not explicitly type of 'string' or for long floats that are encased in quotes, making it a string that will be truncated/rounded when cast to a float.  So, further logic must be completed to test for the case.  Take the following example:

<?php

if (!function_exists("test_float")) {

    function test_float($test) {

        if (!is_scalar($test)) {return false;}

        $type = gettype($test);

        if ($type === "float") {
            return true;
        } else {
            return preg_match("/^\\d+\\.\\d+$/", $test) === 1;
        }

    }

}

$test = "3.14159265358979323846264338g32795";

var_dump($test);
var_dump((float)$test);
var_dump($test == (string)(float)$test);
var_dump(test_float($test));

?>

Will produce (32-bit):

string(34) "3.14159265358979323846264338g32795"
float(3.1415926535898)
bool(false)
bool(false)

So far, so good, right?  Yeah, but it's misleading, because the string is so long, that when it's converted to a float, it won't be equivalent to the comparison of the value being cast back into a string .  So the aforementioned short function works.  Look at this next example:

<?php

$test = 3.1415926535897932384626433832795;

var_dump($test);
var_dump((float)$test);
var_dump($test == (string)(float)$test);
var_dump(test_float($test));

?>

Will produce (32-bit):

float(3.1415926535898)
float(3.1415926535898)
bool(false)
bool(true)

Why is it not working now, but the value is truly a float?  Same reasoning as mentioned before.  The float is so long that it's truncated/rounded and doesn't match the comparison being done with the short-hand function.

So, as you can see, more logic should be applied to the variable you're testing.

Yet another regular expression for float in real life:

<?php
function isTrueFloat($val)
{
    $pattern = '/^[-+]?(((\\\\d+)\\\\.?(\\\\d+)?)|\\\\.\\\\d+)([eE]?[+-]?\\\\d+)?$/';

    return (!is_bool($val) && (is_float($val) || preg_match($pattern, trim($val))));
}
?>

// Matches:
1, -1, 1.0, -1.0, '1', '-1', '1.0', '-1.0', '2.1', '0', 0, ' 0 ', ' 0.1 ', ' -0.0 ', -0.0, 3., '-3.', '.27', .27, '-0', '+4', '1e2', '+1353.0316547', '13213.032468e-13465', '-8E+3', '-1354.98879e+37436'

// Non-matches:
false, true, '', '-', '.a', '-1.a', '.a', '.', '-.', '1+', '1.3+', 'a1', 'e.e', '-e-4', 'e2', '8e', '3,25'