diff --git a/producer_pool.go b/producer_pool.go
new file mode 100644
index 00000000..e32632ac
--- /dev/null
+++ b/producer_pool.go
@@ -0,0 +1,99 @@
+package nsq
+
+import (
+	"errors"
+	"sync/atomic"
+)
+
+// ProducerPool provides a round robin publish to a pool of producers with built in retry
+//
+// This includes a transparent retry when a publish fails, and a backoff when
+// encountering errors
+type ProducerPool struct {
+	Producers   []*Producer
+	MaxAttempts int
+	next        uint32
+}
+
+func NewProducerPool(addrs []string, cfg *Config) (*ProducerPool, error) {
+	p := &ProducerPool{
+		Producers: make([]*Producer, len(addrs)),
+	}
+	for i, a := range addrs {
+		np, err := NewProducer(a, cfg)
+		if err != nil {
+			return nil, err
+		}
+		p.Producers[i] = np
+	}
+	return p, nil
+}
+
+// Publish does a round-robin publish, if a publish fails it will retry up to 3 attempts
+// and will publish sequentially to the next sequential items in the pool
+func (p *ProducerPool) Publish(topic string, body []byte) error {
+	n := atomic.AddUint32(&p.next, 1)
+	l := len(p.Producers)
+	var err error
+
+	for attempt := 0; (attempt <= p.MaxAttempts || p.MaxAttempts <= 0) && attempt < l; attempt++ {
+		producer := p.Producers[(int(n)+attempt-1)%l]
+		err = producer.Publish(topic, body)
+		if err == nil {
+			return nil
+		}
+		producer.log(LogLevelInfo, "(%s) publish error - %s", producer.conn.String(), err)
+	}
+	return err
+}
+
+// PublishAsync does a round-robin asynchronous publish. If a publish fails it will enter a retry loop
+// to try the next sequential publisher in the pool. The retry loop will be executed for maxPublishRetries
+// or the number of publishers in the pool, whichever is smaller. If the message fails to publish after
+// exhausting all attempts, the failed transaction will be written to doneChan for handling.
+// If successful, the transaction is written to doneChan for handling.
+func (p *ProducerPool) PublishAsync(topic string, body []byte, doneChan chan *ProducerTransaction, args ...interface{}) error {
+	if len(p.Producers) < 1 {
+		return errors.New("no producers")
+	}
+
+	n := int(atomic.AddUint32(&p.next, 1))
+
+	maxAttempts := len(p.Producers)
+	if maxAttempts > p.MaxAttempts {
+		maxAttempts = p.MaxAttempts
+	}
+
+	go func(n int) {
+		ch := make(chan *ProducerTransaction, 1)
+		defer close(ch)
+
+		for attempt := 0; attempt < maxAttempts; attempt++ {
+
+			isLastAttempt := attempt+1 == maxAttempts
+
+			index := (n + attempt - 1) % len(p.Producers)
+			producer := p.Producers[index]
+
+			if err := producer.PublishAsync(topic, body, ch, args...); err != nil {
+				if isLastAttempt {
+					doneChan <- &ProducerTransaction{Error: err, Args: args}
+					break
+				}
+
+				producer.log(LogLevelInfo, "(%s) PublishAsync error - %s", producer.conn.String(), err)
+				continue
+			}
+
+			transaction := <-ch
+			if transaction.Error != nil && !isLastAttempt {
+				producer.log(LogLevelInfo, "(%s) PublishAsync error - %s", producer.conn.String(), transaction.Error)
+				continue
+			}
+			doneChan <- transaction
+			break
+		}
+	}(n)
+
+	return nil
+}