Don't know if you're still trying to get it working, but I thought I'd throw out what worked for me. For starters I had to set the echo pin as a pullup input:
pinMode(ECHO_PIN,INPUT_PULLUP);
Originally I was getting 0cm back almost every ping; that got me a reading maybe 4 or 5 out of 6 pings, with intermittent 0cm readings. But the readings were all over the place. I watched the trigger and echo pins on a scope and it was fairly regularly not responding to the trigger. I thought maybe the timings were off slightly on either my arduino or the SR04T, so I edited the ping_trigger() function in newping.cpp, changing the delays from 4 and 10 to 6 and 12:
*_triggerOutput &= ~_triggerBit; // Set the trigger pin low, should already be low, but this will make sure it is.
delayMicroseconds(6); // Wait for pin to go low.
*_triggerOutput |= _triggerBit; // Set trigger pin high, this tells the sensor to send out a ping.
delayMicroseconds(12); // Wait long enough for the sensor to realize the trigger pin is high. Sensor specs say to wait 10uS.
*_triggerOutput &= ~_triggerBit; // Set trigger pin back to low.
Results have been great. Haven't had a missed ping since.
On a side note, I changed the output on ping_in (the only one I use) to float so I could get fractions of an inch. I saw at least one post where somebody requested decimals in ping_cm or ping_in and the author of NewPing responded that the sensors aren't capable of sub-centimeter accuracy, so the extra precision was not useful. But in my testing, my sensor results tend to stay within about 0.4-0.5" of each other, at least at sizeable distances, barring the occasional wildly inaccurate reading which is easily detectable. Mine gives a reading that's 15-50" different from the rest about every 15-20 pings, but because they're so far off, it's easy to pick out and discard them.
I left it running for an hour, taking a measurement every second, while pointed at my ceiling from my desk. After discarding any measurements more than 5" above or below the average, the low measurement was 64.87", high was 64.93", and the vast majority of the measurements were between 64.88 and 64.90. Lowering the sensor a bit at a time seems to indicate that the sensor can reliably detect changes greater than 0.1". I'd say a quarter of an inch precision, at least at the above distance, isn't unreasonable to assume.